Skip to content
WE BUILDHUMANOIDS

INTELLIGENCE · /updates

Updates

Releases, research, open-source activity and industry developments across humanoid robotics. Every entry links to its original source and is reviewed before publishing.

72 updates · newest first

1 UPDATE

OPEN SOURCE

Unitree Robotics lists G1_Dex1_Arrange_Surgical_Kits dataset on Hugging Face

Unitree Robotics has a dataset named G1_Dex1_Arrange_Surgical_Kits on Hugging Face. It sits in a collection called UnifoLM_G1_Dex1_Dataset, which the page shows with 81 items. The page tags the dataset with an Apache licence and says it was created with the LeRobot framework. The card describes 100 episodes recorded at 30 fps, with 297,368 rows and a total file size of 24.5 GB. All cameras are AV1-encoded at 480×640×3. The video streams cover raw and rectified head stereo cameras, left and right wrist colour cameras, and two infrared cameras on each wrist. Observations include 7-dimensional left and right arm states (three shoulder joints, the elbow and three wrist joints), a 15-dimensional lower body state covering both legs and the waist, and 6-dimensional end-effector poses in both the base frame and the torso frame. The torso pose and a camera pose are also recorded, each as 6 values, along with a flag showing whether the wrist infrared streams are valid. The action space mirrors the arms, lower body and waist, and adds target end-effector poses in base and torso frames. It also includes a 4-dimensional base command (vx, vy, angle_z, height) and 4 gripper control channels: left and right trigger and squeeze. The dataset uses Dex1 grippers, with one channel per side. The card gives the gripper range as 5.5 to 0.0, from open to close. The page lists the Hugging Face task as robotics and the format as parquet.

Unitree G1Unitree Robotics (Hugging Face) (original source)OFFICIAL

2 UPDATES

OPEN SOURCE

Unitree publishes R1-7A teleoperation and data-recording code on GitHub

Unitree Robotics has added a repository called R1-7A to its GitHub organisation. The project is a Python teleoperation framework for a seven-axis robotic arm, built on PICO and XRoboToolkit. It covers single-arm and dual-arm control, recording from several cameras, and replay of recorded trajectories. An operator wears a PICO headset and holds its two controllers, which drive the arm or arms and the Dex1 grippers. The framework tracks the relative pose of each controller, and the left and right arms and grippers are controlled independently. The README's hardware list also names a PICO headset running the XRoboToolkit client and a four-channel USB-to-485 module that connects the host computer, arms and grippers. Grippers can be switched off when none is fitted. All modes share one entry point, scripts/teleop.py, with a flag to choose the right arm, the left arm or both. Adding the record flag stores data in MCAP files: control targets, measured joint states and images from several cameras. A single arm produces 8-dimensional data and a dual-arm setup 16-dimensional data. The repository gives a target rate of 90 Hz for control on the real arm and 30 Hz for image observations. Recorded episodes can be replayed on the real arm, or viewed in MuJoCo simulation, and a mock input mode runs the dual-arm simulation without hardware. The README lists limits. The base must be fixed and the arm supported before a real start-up, and collision planning for two arms is not currently provided. The repository has no LICENSE file at its root, and third-party dependencies keep the licences of their own projects. The section on training and deployment is empty.

GitHub (original source)OFFICIAL
OPEN SOURCE

Six AGIBOT G2 Robots Complete 1,730 Tasks in 24-Hour Shanghai Convenience Store Trial

AGIBOT says that six of its G2 wheeled humanoid robots completed a 24-hour trial at an operating convenience store in Shanghai's Xuhui District, carrying out a combined 1,730 tasks. The trial ran from 10 a.m. on November 2 to 10 a.m. on November 3 and was livestreamed to more than two million viewers. It closed AGIBOT's AI WAVE technology showcase, which began on October 22. AGIBOT reports 683 customer-service interactions, 369 product pickups and deliveries, 147 coffee orders, 126 breakfast servings, 277 shelf inspections and restocking tasks, and 128 table-cleaning tasks. Four of the six robots passed the final assessment, which looked at time on duty, task completion, customer feedback, response times and handling of unexpected situations. Two did not meet all the criteria. Two evaluators also ran practical tests. They ordered regular and sugar-free Coca-Cola to see whether the robots could tell similar-looking products apart, rearranged packs of chewing gum for the robots to put back, and blocked an aisle with a person and a box to check obstacle detection and navigation. According to AGIBOT, the G2 robots used real-time visual information to return misplaced items to shelf positions instead of relying only on preset coordinates. One restocking robot found gaps on shelves and refilled them without a specific prompt, and a cleaning robot told spills apart from countertop patterns and chose suitable tools. The trial also exercised Genie Master, AGIBOT's embodied AI agent framework. Through the Embodied Execution Protocol (EEP), Genie Master links task planning, execution and feedback, and results go back to it for analysis and troubleshooting. AGIBOT says it can keep experience from successful and unsuccessful attempts in long-term memory and turn effective adjustments into training data to refine the robots' action policies. The company plans to keep testing its robots in further roles and real-world environments.

AGIBOT Genie G2AGIBOT (original source)OFFICIAL

1 UPDATE

INDUSTRY

NEURA Robotics, E.ON and enviaM-Gruppe launch partnership to bring Physical AI to critical energy infrastructure

NEURA Robotics, E.ON and the enviaM Group announced on 5 November 2026 that they will develop Physical AI for power-grid operations and move it step by step into practice. The partners name inspection and monitoring of technical assets as target applications, along with support for work that is hazardous, physically demanding or repetitive. The work builds on Neuraverse, which NEURA describes as its open Physical AI platform. Skills are to be trained first at TUM RoboGym powered by NEURA, which NEURA and the Technical University of Munich opened at LabCampus at Munich Airport in October 2026. NEURA says the centre covers more than 2,300 square metres and that companies from more than ten industries already train robots there. The initial tasks are inspections, monitoring of critical assets such as substations, and support for site security; tasks in hazardous environments will also be assessed. NEURA says its NEURA Gym combines human demonstrations by grid specialists, real-world robot data, and simulation with a Data Amplifier, and that skills are then validated on real robots. Training uses Neuron AI, which NEURA calls its multimodal foundation model for Physical AI, combining 3D vision, touch, audio, robot state, language and memory. NEURA states that humanoid robots such as its 4NE1 will in future be able to navigate grid facilities, with their stairs, doors and control cabinets, without modifications. Neuraverse is meant to make a skill trained in the NEURA Gym reusable on compatible robots and at other sites, and E.ON and enviaM retain control of their own data and trained skills. MITNETZ STROM, the grid subsidiary of the enviaM Group, is the pilot partner and will test the first applications in real grid operations. The enviaM Group's CEO, Stephan Lowis, says it expects to gain initial practical experience during 2027.

NEURA Robotics (original source)OFFICIAL

1 UPDATE

INDUSTRY

Churchill Capital Corp XI Shareholders Approve Business Combination with Agility Robotics

Shareholders of Churchill Capital Corp XI, a publicly traded special purpose acquisition company, have approved its business combination with Agility Robotics, the maker of the Digit humanoid robot. Churchill XI announced the result of the special meeting on 2 November 2026, saying that more than 84% of votes cast were in favour and that all related proposals were approved. Agility says its gross capital raise from the transaction and related financings is expected to total approximately $465 million, before transaction expenses. That is made up of approximately $164 million from Churchill XI's trust account, approximately $201 million from a common stock PIPE committed at $10 per share and led by Foxconn, and $100 million of equity funding already received after the combination was announced. Agility says the money is meant to support its business plan and accelerate commercial deployments with enterprise customers. As part of the deal, Churchill XI will move its jurisdiction of incorporation from the Cayman Islands to the State of Delaware, and the continuing entity will be renamed Agility Robotics, Inc. Agility describes Digit as a general-purpose humanoid robot that is commercially deployed, and cites more than 65,000 hours of real-world operation. The transaction and the PIPE financing are expected to close on 4 November 2026, subject to closing conditions. The combined company's common stock and warrants are then expected to begin trading on the Nasdaq Stock Market on 5 November 2026 under the ticker symbols AGLT and AGLTW.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

RESEARCH

HLO-AMP: Hierarchical Loco-Manipulation from Object-Aware Motion Priors

A paper on arXiv by Jean-Pierre Sleiman, Bilal Hammoud, He Li, Jonathan Foster, Sylvain Bertrand, Jessica K. Hodgins and Farbod Farshidian describes HLO-AMP, a hierarchical approach to humanoid loco-manipulation. It was submitted on 3 November 2026 and has been accepted at the Conference on Robot Learning (CoRL 2026). The authors start from a common set-up in which a task-agnostic whole-body controller sits under a downstream task policy. They note that such controllers are often trained with manual reward shaping, which gives visibly machine-like motion, and without the manipulated object in the scene. Object dynamics are approximated with randomised force disturbances at the end-effectors. The paper says this proxy does not reproduce the multi-contact wrench that a real payload applies across the arms and torso, and that the gap becomes critical near the platform's payload limit. Their alternative is to train the low-level controller as an object-aware motion prior. It is a single policy trained by motion imitation on retargeted motion-capture references covering the core motion primitives of the downstream task, with the object physically present in the scene. The paper says this stage inherits human-like coordination from the references and absorbs the contact wrench at its source, while keeping the compact task-space command interface of command-tracking controllers. A reference-free high-level policy is trained on top. Its action channel is anchored to the empirical envelope of the low-level training data, so it can extend the priors to a long-horizon, multi-stage task while structurally avoiding queries that would push the low level off-distribution. The framework is instantiated on a Unitree G1 for bimanual box pick-and-place. The authors report that the same deployed policy generalises across object masses and dimensions and across initial and goal poses, and that it recovers reactively from intentional drops during carrying.

Unitree G1arXiv (original source)REVIEWED

1 UPDATE

INDUSTRY

NEURA Robotics shows 4NE1 Wheels running a laundry cycle live at GTC Berlin

NEURA Robotics says its humanoid 4NE1 Wheels completed a full laundry cycle autonomously, repeatedly, throughout NVIDIA GTC Berlin 2026, held from October 20 to 22 at Station-Berlin. The demonstration ran at Booth 3031 in Hall 3 and was also shown online. The company describes it as an unedited run, in contrast to the cut and sped-up videos common in robotics. According to NEURA, the robot sorted light and dark garments, loaded a washing machine and closed its door, then unloaded the washed garments, moved them to a dryer and closed that door. The sequence took around eleven minutes without human intervention. The company shortened the wash and dry cycles of the BSH appliances to about ten seconds, and says all other conditions were real. For this first public run NEURA put reliability and safety around visitors ahead of speed, and says speed will be tuned in later model updates. The robot ran on Neuron AI, the multimodal foundation model NEURA introduced on October 14. NEURA says its Perception-Language-Action architecture combines 3D vision, touch, force, audio, language and the robot's own state with memory. At the booth, a chest camera mapped the workspace and cameras in the hands guided grasps. Tactile sensing took over inside the washing drum, where sight is blocked, and sound helped confirm events such as a door latching. NEURA says the model learns which sense to rely on at each stage. A Thursday keynote by NEURA ended with 4NE1 Gen 4, which the company launched the same day at the opening of the TUM RoboGym in Munich. The robot walked from the keynote stage to the booth. NEURA says it is designed for series production, is 20 percent lighter than its predecessor, has a slimmer body with a textile outer skin and runs on Neuron AI.

NEURA Robotics (original source)OFFICIAL

1 UPDATE

INDUSTRY

Agility Robotics Expects About $465 Million Gross Raise as Churchill Capital Corp XI Merger Nears Closing

Agility Robotics and Churchill Capital Corp XI, a special purpose acquisition company listed on Nasdaq, announced on 30 October 2026 that Agility's gross capital raise from the proposed business combination and related financings is expected to be approximately $465 million, before transaction expenses. The figure follows the expiry of Churchill XI's redemption deadline. According to the announcement, the total is made up of approximately $164 million from Churchill XI's trust account, approximately $201 million from a common stock PIPE committed at $10 per share and led by Foxconn, and $100 million of equity funding already received after the business combination was announced. The companies say the combination is expected to close on 4 November 2026, subject to approval by Churchill XI shareholders at an extraordinary general meeting on 2 November 2026 and to customary closing conditions. The combined company would operate as Agility Robotics, with its common stock expected to begin trading on Nasdaq under the ticker symbol AGLT on 5 November 2026. All existing Agility shareholders will roll their equity into the combined company and be subject to a 180-day lock-up at closing. On the product side, Agility says Digit 5, unveiled on 15 September 2026, is designed to work safely alongside people and support a broader range of industrial workflows. It says it is on track to begin deliveries to early-access customers in the first half of 2027, with general availability targeted for the end of 2027. The company cites an expansion with GXO to deploy Digit at two additional facilities, an AGCO manufacturing pilot expected to begin in 2027, and more than 30 prospective customers in its pipeline. It has previously disclosed more than $300 million in committed multi-year orders for Digit 5, which it says is potential value subject to contractual milestones, product features and specifications, and not current-period revenue.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

AGIBOT unveils GO-3 Preview, an embodied foundation model for learning across the full robot lifecycle

AGIBOT has unveiled GO-3 Preview, an embodied foundation model that it says supports learning at scale through pre-training, post-training and post-deployment. The company reports first place on three embodied AI benchmarks, RoboDojo, RoboTwin and RoboColiseum, and says the model will be fully open-sourced by the end of 2026. The model is built on WRAM, the World Reasoning Action Model, an embodied-native architecture that joins task understanding, future-state prediction and action generation. Three components carry this: CoAE, a control-oriented autoencoder for visual representations; SVP, which predicts a complete future state in one step; and KASO, which selects future predictions more consistent with the actions taken in training. Pre-training draws on human first-person videos and robot trajectories. AGIBOT says that in zero-shot settings the model can apply learned capabilities to new tasks without task-specific supervised fine-tuning on physical robots. For post-training, AGIBOT built Genie Evolver, which combines a fleet of 100 physical robots with physics simulation and world models. Its Collective Experience Learning component aggregates execution experience across multiple robots in real-world environments. AGIBOT illustrates the approach with tabletop curling, a task in which a robot must control both the force and the angle of each shot. The company says that after just a few hours of practice, combining real-world interaction with world-model learning, the robot achieved a higher win rate than typical human players. After deployment, the Genie Master agent framework uses the Embodied Execution Protocol, which standardises interfaces for capabilities, guidance, feedback and continuity. AGIBOT says its Genie G2 improved across multiple complex manipulation tasks in a real convenience store without human demonstrations or manual tuning.

AGIBOT Genie G2AGIBOT (original source)OFFICIAL

2 UPDATES

RESEARCH

FORGE: Feasible Online Reduced-Order Reference Generation and Whole-Body Execution for Agile Humanoid Locomotion

Adrian B. Ghansah, Yujin An, Jeeseop Kim and Soon-Jo Chung have posted a paper on arXiv describing FORGE, a two-stage learning framework for agile humanoid locomotion. It links online reference generation from a reduced-order model (ROM) with a whole-body tracking controller. The authors submitted it on 27 October 2026, and it has been accepted to the 2026 IEEE-RAS 25th International Conference on Humanoid Robots (Humanoids). In the first stage, the authors train a reference generator that takes commands as input and produces whole-body references. It is built on single rigid body dynamics combined with the full kinematics of the humanoid. In the second stage, the reference generator is frozen and rolled out online, and the references it produces are used to train a whole-body tracking policy on the full-order dynamics model. The authors say that learning a kino-dynamically consistent reference generator in the simplified ROM space lets them produce references across a broad command space and across several locomotion strategies, within a single framework. They demonstrate the method on a Unitree G1 humanoid robot. Reported results include longitudinal velocities from -2.0 m/s to 2.8 m/s, lateral velocities up to +-1.0 m/s and angular velocities up to +-3.0 rad/s, as well as one-legged hopping and continuous transitions between commands. According to the abstract, one reference policy and one whole-body tracking policy cover all of these behaviours.

Unitree G1arXiv (original source)REVIEWED
RESEARCH

Humanoids 2026 paper adopts stochastic motion tokenization for whole-body control on the Unitree G1

Michael Drolet, Oleg Arenz and Jan Peters have posted a paper on arXiv, submitted on 27 October 2026, that proposes controlling humanoids through compact motion representations instead of velocity commands. The paper has been accepted to the 2026 IEEE-RAS International Conference on Humanoid Robots (Humanoids 2026). The authors start from two limitations. Velocity-commanded reinforcement learning policies work well for quadrupeds, but they are a limited interface for a humanoid that has to perform complex whole-body motions. Tracking policies for such motions also tend to know little about the desired future motion, which makes it hard for them to deviate in a controlled way to succeed at a long-term task, for example when preparing a transition between behaviours. Their approach has a high-level controller generate concise motion representations from a behaviour label or a text command. These representations are passed straight to a tracking policy, where they serve as a detailed motion reference. The authors say the method handles real-world motions such as dancing, fighting and walking, depending on the conditioning information. The authors adopted a stochastic motion tokenizer for these motion representations. They report that, at matched capacity, it reconstructs motions from the Unitree G1 at a level that exceeds finite scalar quantization (FSQ) and the vector-quantized autoencoder (VQ-VAE). The interface that results is discrete, and on a physical Unitree G1 it conditions whole-body tracking policies that execute generated dance, fight and walk motions lasting 16.8 s each, plus a 35 s fight-to-dance sequence.

Unitree G1arXiv (original source)REVIEWED

3 UPDATES

INDUSTRY

GXO and Agility Robotics Expand Partnership to Deploy Digit at Two More U.S. Sites

GXO Logistics and Agility Robotics announced on 26 October 2026 that they are expanding their partnership, with Agility's Digit humanoid robot going to GXO sites in North Carolina and Tennessee. Under the expanded agreement, Digit will support e-commerce fulfilment at GXO's Statesville and Lebanon sites, taking on highly repetitive material-handling tasks. Deployments are expected to begin in early 2027 using Digit 4, and are to move to Digit 5 once that robot becomes commercially available. The expansion builds on a multi-year Robots-as-a-Service agreement signed in June 2024, which followed a proof-of-concept pilot at GXO's Flowery Branch site in late 2023. Under that agreement GXO deployed Digit robots together with Agility Arc, Agility's cloud-based platform for managing robot fleets and integrating them into customer operations. At Flowery Branch, Digit works alongside existing automation, moving totes and placing them onto conveyors. Agility says Digit has moved more than 100,000 totes there. Agility also says Digit has accumulated more than 65,000 hours of real-world operation across its commercial operations and customer programmes, and that this experience informed Digit 5. The company unveiled that robot on 15 September 2026 and describes it as designed for cooperatively safe work in close proximity to people.

Agility Digit 5Agility Robotics (original source)OFFICIAL
RELEASE

Agility Robotics Reports Expanded GXO Deployment and New Manufacturing Customers for Digit

Agility Robotics said on 26 October 2026 that demand for its Digit humanoid is growing. The company reported an expanded relationship with GXO, three additional manufacturing customers and a pipeline of more than 30 prospective customers. CEO Peggy Johnson was due to discuss the progress on an investor call the same day. GXO announced a commercial agreement to deploy Digit at two more U.S. facilities, in Statesville, North Carolina, and Lebanon, Tennessee. Agility says deployments are expected to begin in early 2027 using Digit 4, with a transition to Digit 5 planned as it becomes commercially available. The company says Digit has moved more than 100,000 totes at GXO's Georgia facility under a multi-year agreement that began in June 2024. The three new manufacturing customers are unnamed: a major global automaker, a developer of automotive technology and components, and a global consumer products manufacturer. Agility says all three are in paid programmes focused on specific manufacturing needs. It also lists AGCO, which plans a Digit pilot in Hesston, Kansas, expected to begin in 2027, Toyota Motor Manufacturing Canada, Schaeffler and Mercado Libre. The more than 30 prospective customers are in active commercial discussions or in its Customer Acceleration Program, which lets them evaluate Digit against specific operational requirements. Among them are two potential subcontractor engagements supporting a branch of the U.S. Armed Services. Agility describes these discussions as early, with no agreements signed. On Digit 5, Agility says it has previously disclosed more than $300 million in committed multi-year orders, which it describes as potential value subject to contractual milestones and not current-period revenue. Early customer access is expected in the first half of 2027, with general availability targeted for the end of 2027.

Agility Digit 5Agility Robotics (original source)OFFICIAL
OPEN SOURCE

AGIBOT Launches Genie Studio 2.0 to Bring Robots into Real Workplaces

On 26 October 2026 AGIBOT launched Genie Studio 2.0, a platform for developing, deploying and operating embodied AI applications. The company says it can support initial deployment and validation in a new task environment in as little as seven days. It is split into three modules: Flywheel for model development, Agent for task orchestration and Monitor for fleet operations. Flywheel links data collection, model training, simulation testing and deployment on physical robots, and feeds failure and anomaly data back into the next training cycle. Customers can train visual perception, vision-language-action (VLA) and reinforcement learning models, and the platform gives access to AGIBOT's GO-2 embodied foundation model. Genie Sim can generate simulation data for situations that are rare, costly or risky to capture on hardware. AGIBOT says its Genie Sim Benchmark covers four evaluation dimensions and 42 task categories, and that simulation results differ from physical-robot performance by less than 10%. Agent lets teams build workflows through a visual drag-and-drop interface, combining navigation, visual localisation, model inference and decision logic, including calls to VLA and reinforcement learning models. Workflows can be tested in simulation for paths, collisions and robot poses, then saved as templates for reuse. Monitor gives a central view of each robot's status and task progress, records faults, raises alerts and allows remote intervention. AGIBOT reports use in manufacturing and rail transit. On a pilot validation line at PIA Automation, three AGIBOT G2 robots loaded and assembled components for automotive battery management system controllers; AGIBOT states that success rates stayed above 99.9% with cycle times of 13 seconds or less per operation. It also says the platform supported more than 60 devices at Guangzhou Metro stations.

AGIBOT Genie G2AGIBOT (original source)OFFICIAL

1 UPDATE

RESEARCH

HumanoidNDM: Closing the Training-Deployment Gap of Motion Tracking via Humanoid Neural Dynamics Model

A paper posted to arXiv on 25 October 2026 by Chao Chen, Zhenrong Wang, Jilong Wang, Yusen Feng, Jiangran Lyu, Li Yi and He Wang describes HumanoidNDM, an online fine-tuning framework for humanoid motion tracking. The authors start from the observation that robots in real deployment meet dynamics gaps, caused either by extra physical disturbances or by inaccurate simulation modelling, and that these gaps open a significant difference between training and deployment. They argue that using real-world data to fine-tune policies is the last mile for robust deployment. The framework has three stages. For data collection, the authors introduce a low-frequency reference perturbation method with an adaptive intensity mechanism, which they say balances deployment feasibility against exploration efficiency. For model training, they train a history-based, proprioceptive neural dynamics model in Euclidean space. It incorporates differentiable forward kinematics, which the authors say removes the reliance on costly motion capture systems. In the final stage, the policy is fine-tuned directly through backpropagated model gradients, which the authors say gives high data efficiency. The paper reports physical validation on the Unitree G1 robot and says HumanoidNDM enables rapid real-world adaptation to various perturbations. The authors describe it as the first online, motion-capture-free fine-tuning framework for humanoid tracking policies. The arXiv page notes that the paper has been accepted to CoRL 2026.

Unitree G1arXiv (original source)REVIEWED

1 UPDATE

OPEN SOURCE

AGIBOT open-sources Physical Reasoning dataset as AGIBOT WORLD 2026 Release 4

AGIBOT announced on 22 October 2026 that it has open-sourced the "Physical Reasoning" dataset, the fourth release of AGIBOT WORLD 2026. The company says this is the first release in the initiative to move the task design away from manipulation difficulty and towards the complexity of decisions, so that robots can choose and carry out actions as conditions change. The dataset records how the same goal is handled under different states and constraints, with more than 10 categories of logical reasoning scenarios. AGIBOT names four factors that vary: object quantity, occlusion, spatial occupancy and task progress. Examples in the release include restocking a hotel-room minibar and retrieving a key that may sit behind a box. It provides high-level question-and-answer data together with low-level action trajectories, which AGIBOT says allows high-level decision models and low-level action policies to be trained together. The dataset addresses nine reasoning abilities, among them goal-directed search, long-horizon planning, memory and state maintenance, and subtask ordering. Tasks come in three difficulty levels: Easy, Medium and Hard. Execution outcomes are annotated, and successful and erroneous samples carry equal weight, which AGIBOT presents as material for studying how systems adjust after deviations. The annotations include a high-level instruction, a task frame, subtask segments and action steps. At action level, a Dense Prompt describes the target object, manipulation region, bimanual role assignment, motion process, gripper state and expected outcome. Bounding-box, point and trajectory-line annotations support visual-language grounding, and "success segments" and "adversarial segments" help assess task completion. AGIBOT says the dataset has already supported its upcoming agent framework and the training of its next-generation foundation model, with further progress to be announced.

AGIBOT (original source)OFFICIAL

2 UPDATES

INDUSTRY

NEURA Robotics and TUM open RoboGym in Munich and launch 4NE1 Gen 4

On 22 October 2026 NEURA Robotics and the Technical University of Munich (TUM) opened the TUM RoboGym powered by NEURA at LabCampus near Munich Airport. The centre covers more than 2,300 square meters and was built with TUM's Munich Institute of Robotics and Machine Intelligence (MIRMI). To mark the opening, NEURA also launched 4NE1 Gen 4, the new generation of its humanoid robot. Companies bring concrete tasks and process knowledge, and researchers study how robots can learn those tasks faster and perform them more reliably. NEURA says partners have been training workflows there since September, with companies from more than ten industries involved. Humanoids, mobile assistance robots such as MiPA and cognitive robot arms are available as form factors. The RoboGym generates training data in three ways: real-world robot data captured by sensors, human demonstrations recorded with sensor-equipped data suits, motion capture and teleoperation, and simulation with a Data Amplifier that adds variations before skills are validated on real robots. NEURA says 4NE1 Gen 4 is designed for serial production and is 20 percent lighter than the previous generation. It has a slimmer body with a textile outer skin, advanced hands and Smart Limbs, which build reflexes into each limb. It runs on Neuron AI, which NEURA describes as a Perception-Language-Action model that also takes touch, force, sound and the robot's own state as input alongside vision. NEURA says the robot is already in operation at the TUM RoboGym. The TUM RoboGym is a joint initiative of NEURA Robotics and TUM that is initially set up for three years. In total, NEURA and TUM are investing around EUR 17 million in it, of which NEURA contributes approximately EUR 11 million. NEURA describes Munich as the start of an international network of NEURA Gyms, with further locations being established in Europe, the United States and China.

NEURA Robotics (original source)OFFICIAL
RELEASE

Agility Robotics Expands NVIDIA Collaboration as It Prepares Digit 5 for Commercial Scale

Agility Robotics announced on 22 October 2026 that Digit 5 will standardise on NVIDIA IGX Thor for its onboard compute. The company says the change widens its existing collaboration with NVIDIA as it readies the humanoid for broader commercial deployment. The new architecture brings Digit 5's onboard compute together on NVIDIA IGX Thor modules. According to Agility, NVIDIA becomes the exclusive provider of onboard compute for the general availability version of Digit 5. The move builds on the companies' earlier collaboration around the NVIDIA Halos safety platform for physical AI. Agility says it is integrating NVIDIA IGX Thor and further NVIDIA Halos models into the proprietary safe human detection system it uses on Digit. It describes these integrations as intended to extend Digit's current perception capabilities, support connections to a broader ecosystem and aid its ongoing certification and deployment work. Agility will also take part in the NVIDIA Halos AI Systems Inspection Lab, which is to help evaluate Digit's safety-related software, AI components and cybersecurity protections before final third-party certification. On timing, Agility expects early access to Digit 5 to begin in the first half of 2027. Availability is then to increase over the following six to nine months for manufacturing, warehouse and distribution operators. Separately, in the same announcement, Agility says its planned business combination with Churchill Capital Corp XI is expected to close in the fourth quarter of 2026.

Agility Digit 5Agility Robotics (original source)OFFICIAL

2 UPDATES

INDUSTRY

Humanoid widens its work with NVIDIA on onboard computing for its next robots

Humanoid, the UK-based AI and robotics company, announced on 21 October 2026 that it is expanding its collaboration with NVIDIA. The two companies are developing a high-performance robotic edge runtime for Humanoid's next-generation robots, with the aim of running more sophisticated AI directly onboard. Humanoid says it is integrating NVIDIA Jetson Thor, NVIDIA Isaac OS, NVIDIA Holoscan Sensor Bridge and the Holoscan SDK. Jetson Thor is the onboard compute. Humanoid reports that moving to it let the company consolidate two onboard computers into one system while keeping the same computing capacity, with possible extra headroom later. It adds that the unified CPU and GPU memory of Jetson Thor cuts unnecessary data movement and latency, and simplifies how workloads share compute. The Holoscan Sensor Bridge is meant to shorten the path from sensors to AI models. According to Humanoid, camera data can be placed directly into GPU-accessible memory, which reduces data copies and CPU load and allows lower-latency processing across perception, inference and control. Isaac OS is described as an open, deterministic runtime framework that provides a GPU-native runtime for coordinating several robotics workloads. Humanoid says Multi-Gigabit Ethernet provides up to 100Gb of bandwidth for sensor data and command and control signals. The company also says it has used NVIDIA Isaac Lab and Isaac Sim since its early days to train and simulate models before deploying them on physical robots, which gives it a connected workflow from training and simulation through to real-world operation. Humanoid says the companies will keep working together as it advances its robotics platform and prepares its next generation of robots.

Humanoid (original source)OFFICIAL
RELEASE

Agility Robotics Plans New Pittsburgh Facility for Digit Software, AI and Service

Agility Robotics announced on 21 October 2026 that it plans a new 50,000-square-foot facility in Pittsburgh. The company says the site is meant to support its expanding customer base and a growing number of commercial Digit deployments, and that it will add approximately 35,000 square feet to Agility's existing Pittsburgh footprint. Agility says the site will be its centre of excellence for software and AI, concentrating on Digit skill development and software enablement, as well as a major hub for deployment and support teams. Those teams are responsible for supporting Digit fleets in customer operations, deploying new robots and workflows, validating customer applications, developing and validating new skills, and carrying out complex service and repair. Dedicated space will let Agility and its customers validate workflows through proof-of-technology evaluations, while teams develop and test new Digit skills in both simulation and physical environments. Pittsburgh will also become Agility's East Coast service centre, complementing existing service capabilities at RoboFab in Salem, Oregon. Agility describes RoboFab as its manufacturing and operations centre, and its Fremont, California, facility as focused on hardware and physical AI development. On the customer side, Agility says Digit already operates with GXO, Schaeffler and Toyota Motor Manufacturing Canada, and that it is adding four customers across automotive, consumer goods and agriculture. It also says it has secured more than $300 million in multi-year orders for Digit 5, subject to satisfaction of certain contractual milestones. The company plans to hire software engineers, AI engineers, service technicians, deployment specialists and team leaders in Pittsburgh, and expects first employee access to the facility by the end of the year.

Agility Digit 5Agility Robotics (original source)OFFICIAL

4 UPDATES

RESEARCH

SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model

Qiayuan Liao, Yiyang Shao, Junfeng Long, Kevin Zakka, Roberto Horowitz, Pieter Abbeel and Koushil Sreenath have submitted a paper called SuperMimic to arXiv, dated 20 October 2026. It has been accepted at CoRL 2026 as an oral presentation. The paper addresses the cost of training humanoid whole-body controllers that track large human-motion datasets. The authors say current pipelines rely on model-free reinforcement learning. Some merge per-motion specialist policies through lossy distillation; others train a single policy with cluster-scale compute. SuperMimic instead learns one differentiable world model of the dynamics the motions share, fitted from logged transitions of a standard rigid-body simulator. The world model and the tracking policy are trained together in one loop, and backpropagation-through-time over short imagined rollouts gives the policy analytic gradients. In 3 hours on a single GPU, the authors say SuperMimic exceeds the return of a state-of-the-art motion-tracker baseline trained for 7 days on 128 GPUs, which they describe as a reduction of more than 1000 times in GPU-hours. On the benchmark of 129,514 motions it fails on 25, against 179 failures for the baseline. After 9.8 hours on two GPUs it reaches zero failures. On scaling, the authors report that from 1 to 264 hours of motion the world model's prediction error stays stable, and the GPU-hours needed to reach a target return grow slowly. They also state that the policy transfers to a physical Unitree G1 across diverse and acrobatic motions.

Unitree G1arXiv (original source)REVIEWED
RELEASE

Isaac Lab 3.0 reaches general availability with Newton and kit-less workflows

Isaac Lab 3.0 is now generally available, according to the release page in the isaac-sim/IsaacLab repository. The page calls this the official Isaac Lab 3.0 release and the recommended baseline for new projects and for migrations from Isaac Lab 2.x. Isaac Lab 3.0 introduces a factory-based, multi-backend architecture. Assets, sensors, scenes and environments are backend-neutral and dispatch to whichever implementation is selected, so a supported task can switch physics, rendering or visualization without rewriting its logic. The release page adds that supported combinations remain task-specific. Newton, a Warp-native physics backend, offers MuJoCo-Warp, VBD and MPM solver paths. The release page also lists OV Libraries, which provide OVPhysX for PhysX and OVRTX for RTX rendering as standalone libraries. Isaac Sim is still available for workflows that need the full PhysX and RTX stack, the ROS bridge, XR, livestreaming or Kit tooling, but it is now fully optional. On the data side, asset and sensor data use a ProxyArray type that gives zero-copy access to both Warp and PyTorch. A single isaaclab command handles training and playback across RSL-RL, RL-Games, skrl, Stable-Baselines3, RLinf and TorchRL. For deployment, the release lists LEAPP policy export, Isaac Capture teleoperation integration, and guidance for moving policies between PhysX and Newton in both directions. The release targets Isaac Sim 6.2, Python 3.12, PyTorch 2.11 with CUDA 13.0, NVIDIA Warp 1.17, Newton 1.6.1 and RSL-RL 5.5.1. Migrating projects should note that quaternion order changes from (w, x, y, z) in Isaac Lab 2.x to (x, y, z, w), and the notes advise starting from a fresh Python 3.12 environment. A gallery of 17 browser demos, including a G1 velocity-control scene, runs Newton simulations on the CPU through WebAssembly.

Unitree G1GitHub (original source)OFFICIAL
DEVELOPMENT

MuJoCo Menagerie adds an mjlab variant of the Unitree G1 model

A commit by btaba to Google DeepMind's mujoco_menagerie repository on GitHub adds a new file, g1_mjlab.xml, to the unitree_g1 model directory. The commit is titled "Add g1_mjlab.xml to unitree_g1 (#343)" and touches two files: the new model file and the directory's changelog. The changelog entry describes the file as an mjlab variant of the G1 model. It says the variant is derived from g1_mjx.xml, another model file named in the same changelog. The entry lists what differs from g1_mjx.xml. The variant uses capsule colliders on the feet and has active self-collisions. Its actuators and solver options have been removed, for programmatic specification. The changelog dates the entry 2026-10-19. The commit adds 320 lines in total, according to the page.

Unitree G1GitHub (original source)REVIEWED
RELEASE

AGCO and Agility Robotics Advance Smart Manufacturing with Humanoid Robot Pilot in Hesston, Kansas, Facility

AGCO and Agility Robotics have signed a commercial agreement for a pilot of Agility's Digit humanoid robot in manufacturing. The companies announced it on 20 October 2026. The U.S. pilot is expected to begin in 2027 and will start at AGCO's production facility in Hesston, Kansas. According to the announcement, Digit will initially support assembly and production work: sorting and stacking parts, and placing totes for storage. Agility says the aim is to take over repetitive and physically demanding tasks, which it expects to improve ergonomic safety and operational efficiency. Agility says Digit learns and adapts to changing workflows. It adds that, when used with Agility Arc, its cloud-based platform for deploying and managing robot fleets, companies can integrate automation without extensive infrastructure changes or costly facility upgrades. Thomas Enghof, AGCO's Director of Manufacturing Technology, described humanoid robotics and AI as "teammates" in building flexible factories, and pointed to strict safety governance. Agility lists Toyota, GXO and Schaeffler among the other companies deploying its humanoid robots. The release also says Agility is preparing to become a public company through a business combination with Churchill Capital Corp XI, which it expects to close in the fourth quarter of 2026.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

RELEASE

Agility Robotics Names Mark Vale as Expected Audit Committee Chair Ahead of Public Listing

Agility Robotics said on 19 October 2026 that Mark Vale, CEO and board member at Transportation Insight and a former president of UPS, is expected to join its board of directors and chair the audit committee. The appointment depends on completion of the company's previously announced business combination with Churchill Capital Corp XI, a Nasdaq-listed company (CCXI). Agility says Vale has more than three decades of experience in logistics, operations and finance. He spent over 30 years at UPS, where he was global president and held financial roles including corporate controller and chief financial officer for international operations. At Transportation Insight he was previously chief financial and administration officer. Agility says the appointment adds experience as it expands customer deployments of Digit, scales manufacturing and prepares to operate as a public company. As audit committee chair, Vale will help oversee financial reporting, internal controls and the company's relationship with its independent auditor. Agility expects him to serve alongside the previously announced directors Merline Saintil, Derek Aberle and Pierre Gentin, as well as CEO Peggy Johnson and co-founder Damion Shelton. Bruce Leak and Matt Ocko have stepped down from the board. Under the proposed business combination between Agility Robotics and Churchill Capital Corp XI, the combined company is expected to operate as Agility Robotics and trade on Nasdaq under the ticker symbol AGLT. The release says the proposed transaction will be submitted to the shareholders of Churchill Capital Corp XI for their consideration. It also describes Vale's expected appointment and the expected board composition as forward-looking statements.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

RELEASE

Agility Robotics Selects Nyobolt as its Ultra-Fast Charging Battery Technology Partner for Digit 5

Agility Robotics announced on 15 October 2026 that Nyobolt is its technology provider for a new battery system. Agility says the technology powers one of the most significant upgrades in Digit 5, a fast-recharge battery system that lets Digit work more than 20 hours in a 24-hour day. According to Agility, Digit 5 will have a new battery system with 90 minutes of runtime and a charge time of 9 minutes. That gives a run-to-charge ratio of 10:1, against 2:1 for Digit 4. Agility says this fast recharge is meant to support more than two shifts in a single day. Agility describes Nyobolt's battery technology as built for demanding industrial conditions, combining fast charging, long cycle life and lower system weight. Nyobolt's co-founder and CEO, Sai Shivareddy, said the technology was chosen because Digit 5 will need more power from a smaller, lighter pack in environments where reliability is essential. Agility Chief Business Officer Daniel Diez called uptime the business case for Digit. Agility expects early access to Digit 5 with Nyobolt's battery technology to begin in the first half of 2027, with availability increasing over the following six to nine months for manufacturing, warehouse and distribution operators. In the forward-looking statements and risk factors of its press release, Agility describes Nyobolt as the exclusive provider of battery technology for Digit 5.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

INDUSTRY

NEURA Robotics Introduces Neuron AI: The Brain of the Physical World

NEURA Robotics Introduces Neuron AI Skip to content Products MAiRA LARA MAV MiPA 4NE1 Add-Ons NEURA OmniSensor NEURA SenseKit NEURA Teach NEURA Touch Applications Gluing and Dosing Machine Tending Palletizing Sanding and Polishing Quality Inspection Welding Industries Food and Beverage Logistics Healthcare Metal and Machining Transportation NEURA Gym Neuraverse Company Series C About Us NEURA Group Careers News Blog Fanshop Media Hub Press Contact Us Products MAiRA LARA MAV MiPA 4NE1 Add-Ons

NEURA Robotics (original source)OFFICIAL

2 UPDATES

RELEASE

Isaac Lab publishes browser demos v3 with on-screen joysticks for G1

The Isaac Lab repository on GitHub lists a new pre-release, browser-demos-v3, released on 13 October. The release updates the browser demos for MicroDuck, G1 and ANYmal-D. They gain transparent joysticks inside the viewport, along with compact pause and reset controls. The release notes also describe how MicroDuck behaves. It starts out standing. Dragging the robot applies a fixed horizontal push of 2 N, and dragging the background orbits the camera instead. Releasing the mouse or touch, cancelling it, or losing focus ends the interaction. The notes describe the browser-demos-v3 release as a UI-only update. It reuses the original physics, policy and geometry files, and the notes state that no simulation rebuild is required. The release archive includes the unchanged v1/ and v2/ payloads as well as v3/, and the notes say all three are to be deployed. Compared with the previous compressed archive, it adds about 28 KB. For validation, the notes report that all twelve demos stepped with finite state and no JavaScript errors. They list MicroDuck standing, walking and turning, fixed-force mouse and touch pushes, camera orbit, two simultaneous touch joysticks, cancellation, pause and reset, and the mobile layout as passed. They add that simulation timing stayed near real time in the measured run, and that third-party licence notices are retained.

Unitree G1GitHub (original source)OFFICIAL
RELEASE

Agility Robotics and Churchill Capital Corp XI S-4 Registration Statement Declared Effective by SEC in Connection with Proposed Business Combination

Agility Robotics announced that the US Securities and Exchange Commission declared effective, on October 13, 2026, the joint Form S-4 registration statement it filed with Churchill Capital Corp XI (Nasdaq: CCXI). The filing relates to the proposed business combination between the two companies. Churchill XI has set November 2, 2026 for the extraordinary general meeting at which its shareholders will be asked to approve the combination and related matters. Agility also describes its Digit 5 programme. It says Digit 5, which it recently unveiled, is engineered for cooperative safety so that robots can work in close proximity to people. It says the robot is backed by more than $300 million in committed orders, subject to certain contractual milestones, product features and/or specifications, and is on track for early access customer availability in early 2027. Planned improvements are listed as lifting capacity, battery performance and dexterity. On infrastructure, Agility describes Agility Arc as its cloud-based platform for integrating Digit into customer operations and managing robot fleets across facilities. Its RoboFab facility in Salem, Oregon is designed to produce up to 10,000 robots annually, and a hardware and physical AI development hub in Fremont, California opened in July 2026. The company targets growth in committed customer facility deployments from nine today to up to 25 by year-end 2027. Agility expects the transaction to provide more than $620 million in gross proceeds: approximately $424 million held in Churchill XI's trust account, assuming no redemptions, and approximately $200 million from a common stock PIPE led by Foxconn. The business combination still needs Churchill XI shareholder approval and other customary closing conditions. It is expected to close in November 2026, with the combined company's stock expected to trade on Nasdaq under the ticker AGLT.

Agility Digit 5Agility Robotics (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

AGIBOT Opens Hong Kong Office in Causeway Bay

--> AGIBOT Opens Hong Kong Office in Causeway Bay-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot AGIBOT X1 Full-Stack Open-Source Robot AGIBOT X2 series Fully

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

Unitree lists R1_Dex1_Fold_Towel video dataset on Hugging Face

Unitree Robotics has a dataset called R1_Dex1_Fold_Towel on Hugging Face, under its unitreerobotics name. The page labels the data as video and shows a visualiser entry named Unitree_R1_A7_Dex1. According to the page, the dataset contains 29 episodes at 30 fps, filmed with 4 cameras at 640×480 in mp4v format. The total file size is listed as 1.43 GB, and the page reports 123 downloads in the last month. The dataset belongs to a collection named UnifoLM_R1_Dex1_Dataset, which the page says holds 5 items. The page also carries the note “No dataset card yet”.

Unitree R1Unitree Robotics (Hugging Face) (original source)OFFICIAL

3 UPDATES

INDUSTRY

Boston Dynamics Appoints Rohit Prasad as Chief Executive Officer

Boston Dynamics Appoints Rohit Prasad as Chief Executive Officer | Boston Dynamics Products Explore All Products Back Spot Spot Spot Our agile mobile robot for dynamic sensing and industrial inspection Read More Arm An integrated arm to enable mobile manipulation Read More Payload Additional sensors and hardware to expand Spot’s functionality Read More Extras Add-ons and accessories to keep Spot ready for work Read More Atlas Stretch Orbit Solutions Explore All Solutions Back Inspection

Boston Dynamics (original source)OFFICIAL
RELEASE

Agility Robotics livestreams its Analyst & Investor Day, with Digit 5 on the agenda

Agility Robotics announced on 6 October 2026 that it would host its Analyst & Investor Day that day in New York City, from 12:30 p.m. to 3:00 p.m. ET. The company said the event would be livestreamed through its investor relations website, with the presentation materials posted there and a replay to follow. Agility said the event would feature perspectives from Schaeffler (David Kerr and Courtney Baines), Foxconn (Nelson Hsieh) and AUVSI (Heather Lee), in addition to presentations from its own leadership team. Members of that senior leadership team would discuss Digit 5 and the company's technology platform, including how commercial deployments and customer feedback have shaped the product roadmap and its approach to cooperative safety. Other topics on the list were scaling manufacturing and the humanoid supply chain, safety and standards across the wider robotics landscape, and software, autonomy and Physical AI. On the last of these, Agility said it would cover how advances in intelligence will widen the range of work Digit can perform over time. Agility describes Digit 5, unveiled in September, as a humanoid engineered for cooperatively safe work at scale. It says the robot builds on more than 65,000 hours of real-world operation and on customer feedback. The company also reports more than $300 million in multi-year orders for Digit 5, a figure it says reflects orders as of May 2026 and is subject to contractual milestones. Agility's announcement also says the company plans to become publicly traded through a business combination with Churchill Capital Corp XI, expected to close in the fourth quarter of 2026.

Agility Digit 5Agility Robotics (original source)OFFICIAL
RELEASE

Agility to Host Analyst & Investor Day on October 6, 2026

Agility Robotics announced on September 17, 2026 that it will hold an Analyst & Investor Day in New York City on Tuesday, October 6, starting at 12:30 p.m. ET. The company says the event follows the unveiling of Digit 5, which it describes as its next-generation humanoid robot engineered for cooperatively safe work at scale. According to Agility, its leadership will cover Digit 5 and the company's technology platform, including how customer deployments have shaped the product roadmap and its approach to cooperative safety. Other listed topics are the software and autonomy roadmap, the role of Physical AI in extending Digit's capabilities over time, and plans for scaling deployments and manufacturing. The programme also covers commercial momentum, the operating model and the long-term financial profile, and it includes live question-and-answer sessions. The speakers named are Peggy Johnson, Chief Executive Officer; Jonathan Hurst, co-founder and Chief Robot Officer; Daniel Diez, Chief Business Officer; Jennifer Hunter, Chief Operating Officer; and Michael Beer, Chief Financial Officer, along with other senior leaders. The event was to be streamed live from New York City, with the webcast and presentation materials on Agility's investor relations website. Formal presentations were scheduled to run from 12:30 p.m. ET to approximately 3:00 p.m. ET, and a replay was to be available afterwards. Agility also says it is preparing to become a public company through its previously announced business combination with Churchill Capital Corp XI, which is expected to close in the fourth quarter of 2026.

Agility Digit 5Agility Robotics (original source)OFFICIAL

2 UPDATES

RELEASE

MuJoCo 3.15.0

Version 3.15.0 (October 5, 2026) General 1aa68e1ca .mjz archives written on Windows now always use / separators. b3dd9e617 Actuator dampratio now computes the reflected inertia $m$ at mjModel.qpos0 as the operational-space inertia $(J M^{-1} J^T)^{-1}$ (averaged across force

NEOGitHub (original source)OFFICIAL
RESEARCH

Agility Robotics CEO Peggy Johnson to Join Project Meridian Defence Study

Agility Robotics announced on 5 October 2026 that its CEO, Peggy Johnson, will take part in Project Meridian. The initiative was commissioned by the Department of War and brings together industry executives and former government leaders to examine the future of warfare. Its aim is to inform long-term defence capability development and investment priorities. The MITRE Corporation leads Project Meridian, which will produce strategic recommendations over a 10- to 20-year horizon. The project explores emerging technologies, evolving operational concepts and the capabilities needed to meet future national security challenges. Agility says Johnson will join as an individual consultant, drawing on her technology leadership experience and on the company's work developing and commercially deploying humanoid robots. Agility states that her role is separate from its commercial operations. In a statement, Johnson said she sees an opportunity for humanoid robots to support service personnel by taking on physically demanding and repetitive logistics tasks. She said this could strengthen supply chains and let personnel focus on higher-value work. She also said she intends to bring lessons from real-world commercial deployments to the study. Agility describes its humanoid, Digit, as commercially deployed today. Its announcement presents Johnson's participation as that of an individual consultant to MITRE.

Agility Robotics (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

AGIBOT Partners with Vietnam National Innovation Center(NIC) to Advance Embodied AI Industry and Talent Ecosystem

--> AGIBOT Partners with Vietnam National Innovation Center(NIC) to Advance Embodied AI Industry and Talent Ecosystem-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL

2 UPDATES

INDUSTRY

Inside the Robotics Race

Inside the Robotics Race | Boston Dynamics Products Explore All Products Back Spot Spot Spot Our agile mobile robot for dynamic sensing and industrial inspection Read More Arm An integrated arm to enable mobile manipulation Read More Payload Additional sensors and hardware to expand Spot’s functionality Read More Extras Add-ons and accessories to keep Spot ready for work Read More Atlas Stretch Orbit Solutions Explore All Solutions Back Inspection Inspection Inspection Automate industrial

Boston Dynamics (original source)OFFICIAL
INDUSTRY

Agility and FORT Robotics sign memorandum of understanding on Digit 5 safety

Agility Robotics and FORT Robotics signed a memorandum of understanding on 1 October 2026 for a partnership on the safety infrastructure of Digit 5, which Agility describes as its next-generation humanoid robot for cooperatively safe work. The agreement is meant to deepen a multi-year engagement between the two companies. That engagement started with a single custom hardware component and has grown into an integrated safety architecture in three parts: a safety pendant, on-robot communications, and off-robot interfaces that connect Digit with external safety systems. FORT and Agility developed the off-robot interface jointly. It is designed to extend Agility's cooperative-safety approach beyond the robot, enhancing the onboard safety architecture already planned for Digit 5. Earlier this year Agility engaged FORT to deliver the pendant sled, a custom-built holder for Digit's tablet that integrates emergency stop and enabling-device functionality. Agility says Digit 5, like Digit 4, will be capable of operating autonomously during normal work. The pendant will be used to monitor the robot and give customers a redundant manual override during setup, maintenance or an unexpected situation. Agility says it will complement Digit 5's anticipated onboard Safe Human Detection system, and that Digit 5 will be built on a safety architecture spanning safe human detection, safety cues and safe motion control. The partnership will also establish a broader operating framework for how the two companies work together as Digit 5 begins deployment. Agility says this will include joint hardware development, solutions engineering, regulatory compliance and deployment support. Agility says Digit 5's predecessors have logged more than 65,000 hours of operation, with deployments at Schaeffler, GXO and Toyota Motor Manufacturing Canada. The release describes an MoU and refers to the negotiation and execution of definitive agreements as still to come.

Agility Digit 5Agility Robotics (original source)OFFICIAL

2 UPDATES

PRODUCT

F.02 Decommission

F.02 Decommission FIGURE 03 HELIX INDEX NEWS CAREERS COMPANY F.02 Decommission September 30, 2026 We love Figure It is an incredible robot. F.02 marked many firsts for the company: our first deployment at BMW, the birth of Helix, our first robot doing household chores, and our first logistics deployment. As our F.03 fleet grows, maintaining the F.02 fleet no longer makes sense. Decommissioning a fleet of robots full of custom-built actuators and other intellectual property is no easy task.

Figure 03Figure AI (original source)OFFICIAL
OPEN SOURCE

IDC Data Shows AGIBOT Led Global Humanoid Robot Shipments in H1 2026

--> IDC Data Shows AGIBOT Led Global Humanoid Robot Shipments in H1 2026-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot AGIBOT X1 Full-Stack Open-Source Robot AGIBOT X2

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL

3 UPDATES

DEVELOPMENT

Unitree's G1 description files gain a 31-DoF G1+ model with a pitch-and-yaw neck

Unitree has reworked the G1 robot descriptions in its unitree_ros repository during week 40 of 2026. The latest commit, titled "update G1+", renames the G1+ model from g1_plus to g1_pl: the URDF file becomes g1_pl.urdf, the robot name inside it follows, and the README table and G1+ section use the new name. The commit touches 2 files, with 4 lines added and 4 removed. The README describes g1_pl.urdf as an extension of the g1_29dof_mode_18 model that swaps in a G1+ upper body dated 2026-09-23. The result has 31 degrees of freedom: 12 in the legs, 3 in the waist, 14 in the arms and 2 in the neck. The head moves on separate neck pitch and neck yaw joints, and the model adds revised torso and head meshes, a handle on the torso, and camera and lidar reference frames fixed to the head link. Joint limits change as well. The G1+ model takes its effort and velocity limits from the repository's motor information file; in it, the ankle pitch and roll effort limit rises from 41 to 61.5 N·m and the waist roll and pitch limit from 60 to 90 N·m, while velocity limits stay the same. The other commits grouped in this week's change set add the updated 2-DoF head model and the G1+ model itself, update its motor information and documentation, and revise the G1+ hip motor mapping.

Unitree G1GitHub (original source)OFFICIAL
RELEASE

Agility Robotics Names Three Expected Directors for Planned Public-Company Board

Agility Robotics announced on 29 September 2026 that Merline Saintil, Derek Aberle and Pierre Gentin are expected to join its board of directors once its previously announced business combination with Churchill Capital Corp XI (NASDAQ: CCXI) is completed. They would serve alongside CEO Peggy Johnson and co-founder Damion Shelton, who both remain on the board. Saintil is expected to join the existing board as an independent director immediately, subject to applicable approvals. Aberle and Gentin are expected to join when the proposed combination completes, subject to approvals and closing conditions. Further board members are to be announced later. After the transaction, the combined company is expected to operate as Agility Robotics and trade on Nasdaq under the ticker symbol AGLT. According to Agility, Saintil sits on the boards of Rocket Lab, Symbotic and TD SYNNEX. Aberle spent 17 years at Qualcomm, where he was president from 2014 to 2018, and serves on the board of InterDigital. Gentin most recently was General Counsel of the U.S. Department of Commerce and was previously Chief Legal Officer at McKinsey & Company. On the product side, Agility says its current generation humanoid, Digit 4, is already operating in commercial environments, including at Schaeffler, GXO and Toyota Motor Manufacturing Canada. Agility refers to its recent announcement of Digit 5, which it anticipates will be commercially available in the second half of 2027, and says its focus will then be on scaling its vertically integrated platform. According to the release, the three new directors will together help oversee the company's next phase of growth, as it looks to expand commercial deployments of Digit, scale its manufacturing capabilities and advance the adoption of humanoid robotics across industrial environments.

Agility Digit 5Agility Robotics (original source)OFFICIAL
OPEN SOURCE

AGIBOT and ASD Deploy 100 AGIBOT X2 Humanoid Robots Across 100 Retail Stores

--> AGIBOT and ASD Deploy 100 AGIBOT X2 Humanoid Robots Across 100 Retail Stores-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot AGIBOT X1 Full-Stack Open-Source Robot

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL

1 UPDATE

5 UPDATES

INDUSTRY

PAL Robotics joins ADRF 2026 to strengthen Europe’s AI and robotics ecosystem

PAL Robotics joins ADRF 2026 to strengthen Europe’s AI and robotics ecosystem - PAL Robotics Robots Mobile Manipulators TIAGo Pro TIAGo Pro Revolutionize your work with the next generation of mobile manipulators. Features Tech Specs Software Accessories TIAGo Pro Arm TIAGo Pro Arm The latest addition to the TIAGo Pro Ecosystem. A high-performance, easy to integrate robotic arm for developers that need a truly open platform. Numbers Features Highlights Tech Specs TIAGo TIAGo The mobile

PAL Robotics (original source)OFFICIAL
INDUSTRY

Boston Dynamics Opens Robotics Metaplant Application Center to Train Humanoid Robots for Manufacturing Tasks

Boston Dynamics Opens Robotics Metaplant Application Center to Train Humanoid Robots for Manufacturing Tasks | Boston Dynamics Products Explore All Products Back Spot Spot Spot Our agile mobile robot for dynamic sensing and industrial inspection Read More Arm An integrated arm to enable mobile manipulation Read More Payload Additional sensors and hardware to expand Spot’s functionality Read More Extras Add-ons and accessories to keep Spot ready for work Read More Atlas Stretch Orbit Solutions

Boston Dynamics (original source)OFFICIAL
INDUSTRY

MNACTEC opens its robotics exhibition with the participation of PAL Robotics

MNACTEC opens its robotics exhibition with the participation of PAL Robotics - PAL Robotics Robots Mobile Manipulators TIAGo Pro TIAGo Pro Revolutionize your work with the next generation of mobile manipulators. Features Tech Specs Software Accessories TIAGo Pro Arm TIAGo Pro Arm The latest addition to the TIAGo Pro Ecosystem. A high-performance, easy to integrate robotic arm for developers that need a truly open platform. Numbers Features Highlights Tech Specs TIAGo TIAGo The mobile

PAL Robotics (original source)OFFICIAL
OPEN SOURCE

AGIBOT Demonstrates WITA

--> AGIBOT Demonstrates WITA-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot AGIBOT X1 Full-Stack Open-Source Robot AGIBOT X2 series Fully Intelligent & Agile Robot

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL

1 UPDATE

RESEARCH

EmoPose: Vision-Language Model Guided Emotion-Aware Gesture Generation for Humanoid Robots

Socially competent humanoid robots must communicate affect and intent through gesture as well as speech, yet open-ended interaction must become motion that is both expressive and executable on a specific body. This demands semantic flexibility for contextual social intent while preserving deterministic, embodiment-aware robot control. We present EmoPose, a vision-language model (VLM)-guided framework that bridges this gap through an executable semantic interface. Given language, dialogue

Unitree G1arXiv (original source)REVIEWED

2 UPDATES

RESEARCH

HIGenNTO: Scalable Humanoid Interaction Generation via Noise-Space Trajectory Optimization

Humanoid robots can acquire complex skills by imitating kinematic humanoid motion references, yet reliable references for contact-rich interactions remain difficult to obtain: motion capture deteriorates under occlusion and close physical contact, while retargeting introduces additional contact and geometric inconsistencies. We present HIGenNTO, a framework that synthesizes humanoid-scene interaction motion references by optimizing the initial noise of a pretrained text-conditioned motion

Unitree G1arXiv (original source)REVIEWED
RESEARCH

FootQuery: Future-Touchdown-Guided Retrieval from Depth History for Perceptive Humanoid Locomotion

Humanoid locomotion over complex terrain requires anticipating footholds that may no longer be visible at touchdown. Limited camera coverage and self-occlusion make it necessary to retrieve relevant terrain information from earlier observations. We present FootQuery, a perceptive locomotion framework that queries depth history using each foot's predicted next touchdown. The policy predicts touchdown locations and uncertainty from proprioception and uses these distributions, together with

Unitree G1arXiv (original source)REVIEWED

5 UPDATES

INDUSTRY

Data Sovereignty in Cognitive Robotics: Why Control Over Your Own Data Matters

Data Sovereignty in Robotics: Who Owns Your Machine Data Skip to content Products MAiRA LARA MAV MiPA 4NE1 Add-Ons NEURA OmniSensor NEURA SenseKit NEURA Teach NEURA Touch Applications Gluing and Dosing Machine Tending Palletizing Sanding and Polishing Quality Inspection Welding Industries Food and Beverage Logistics Healthcare Metal and Machining Transportation NEURA Gym Neuraverse Company Series C About Us NEURA Group Careers News Blog Fanshop Media Hub Press Contact Us Products MAiRA LARA

NEURA Robotics (original source)OFFICIAL
RESEARCH

Learning Slope-Adaptive Whole-Body Locomotion for Humanoid Robots in Roofing Construction

Roofing requires workers to coordinate locomotion, balance, and work-related body motions on pitched surfaces, creating a challenging application for humanoid robots. Directly retargeted human demonstrations, however, may preserve motion appearance while placing the robot's feet or hands incorrectly relative to the roof. This study presents a task-semantic scene-grounded framework for learning roofer-style whole-body motions on a Unitree G1. Human demonstrations are captured using a tracking

Unitree G1arXiv (original source)REVIEWED
RESEARCH

Learning Scene-Aware Humanoid Locomotion through 3D Clutter from Immersive Human Demonstrations

While learning from human motions has enabled highly dynamic humanoid skills such as dancing and martial arts in obstacle-free space, traversal through densely cluttered environments remains underexplored. These spaces are three-dimensional and geometrically constrained, requiring scene-aware locomotion that tightly couples whole-body motion with scene geometry for obstacle avoidance. To address these challenges, we present Moving Through Clutter (MTC), a learning-from-demonstration framework

Unitree G1arXiv (original source)REVIEWED
RELEASE

Isaac Lab 3.0 Early Access separates physics, rendering and visualisation backends

NVIDIA released Isaac Lab 3.0 as Early Access on 17 September 2026. The release lets a supported task switch between physics backends, including Newton (with MuJoCo-Warp among its solvers) and PhysX, and run without installing Isaac Sim; full Isaac Sim workflows now require Isaac Sim 6.1. It ships validated policy-transfer examples between PhysX and Newton, one of them rough-terrain locomotion for the Unitree G1. Until general availability the 3.0 branch takes only bug fixes and stability work.

Unitree G1Isaac SimSim-to-realReinforcement LearningGitHub (original source)OFFICIAL
OPEN SOURCE

Figure introduces Helix 2.5, claims zero-shot whole-body tasks in 30 unseen homes

Figure released Helix 2.5, a model pretrained from scratch on its Index human-video dataset and adapted to three whole-body behaviours (tidying, towel folding, bed making). Figure reports these were evaluated in 30 Bay Area homes with no data collected there, and that Index pretraining raised zero-shot success from 9% to 56%.

Figure 03Imitation LearningManipulationWhole-body ControlFigure AI (original source)OFFICIAL

3 UPDATES

RESEARCH

ViLoMan: Learning Visual-Proprioceptive Whole-Body Loco-Manipulation Skills for Humanoid Robots

Humanoid loco-manipulation requires adaptive whole-body coordination to seamlessly integrate locomotion and physical interaction. Despite recent advances, learning autonomous loco-manipulation remains challenging due to the scarcity of diverse, physically executable robot-object interaction data and the difficulty of learning unified whole-body control directly from onboard observations. We present ViLoMan, a scalable framework for autonomous humanoid loco-manipulation. ViLoMan first

Unitree G1arXiv (original source)REVIEWED
RESEARCH

Learning Safe Humanoid Navigation from Reduced Order Models

Research in humanoid robotics has achieved rapid progress in locomotion, and recent results have pushed the boundary on autonomous navigation. We demonstrate that a standard single-stage RL navigation pipeline struggles to scale to multi-level and multi-story terrain, limited by the difficulty of complex humanoid terrain interactions such as stairs. To overcome this challenge, we decompose the navigation problem into two pieces. First, we train a policy operating on the reduced order dynamics

Unitree G1arXiv (original source)REVIEWED
RESEARCH

KINO: A Keyframe Interface for VLM Planning and Whole-Body Control in Humanoid Loco-Manipulation

Humanoid loco-manipulation requires robots to interpret task instructions and scene semantics while executing coordinated whole-body motions. We propose a hierarchical framework that uses motion keyframes as an intermediate representation between Vision-Language Model (VLM) planning and Reinforcement Learning (RL) control. Each keyframe specifies a target whole-body robot pose and, when applicable, an object pose. Given a language instruction, scene observations, and execution feedback, the

Unitree G1arXiv (original source)REVIEWED

1 UPDATE

PRODUCT

Agility unveils Digit 5, its first humanoid designed for cooperatively safe work without barriers

Agility Robotics unveiled Digit 5 on 15 September 2026, describing it as its first humanoid engineered to work in close proximity to people without physical safety barriers. The company cites a new leg design with cycloidal actuators for repeated 50 lb (22.7 kg) lifts and a new battery with a 90-minute runtime, and says it had more than $300 million in multi-year Digit 5 orders as of May 2026, subject to contractual milestones. Early access is expected in the first half of 2027 and general availability by the end of 2027.

Agility Digit 5ActuatorsSafetyAgility Robotics (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

Unitree publishes 44 G1_WBT household-task datasets

In the weeks of 7 and 14 September 2026 Unitree published 44 datasets in a G1_WBT series on Hugging Face (18, then 26). Their names describe household tasks such as making a bed, loading a dishwasher, putting clothes into a washing machine or pushing a wheelchair, in two groups labelled “Brainco” and “Inspire”.

Unitree G1ManipulationImitation LearningDexterous HandsUnitree Robotics (Hugging Face) (original source)OFFICIAL

4 UPDATES

OPEN SOURCE

Unitree releases UnifoLM-ER-Flow, a vision-language model that predicts scene changes alongside robot actions

Unitree Robotics has published UnifoLM-ER-Flow on Hugging Face, a successor to its UnifoLM-ER-1 model. According to the model card, it adds two capabilities to its predecessor: anticipating how a scene is about to change during an interaction, and expressing robot actions as discrete tokens, both handled inside one vision-language model. The weights are released under the Apache 2.0 licence, and the card gives a model size of 4B parameters. To anticipate what an action will change, Unitree isolates the moving parts of a scene with optical flow and compresses them with a VQ-VAE into short token sequences of a set length. Given the current camera image and either a task description or an action, the model then predicts these tokens for the regions that are about to move, which keeps its attention on the objects being handled and on the effects of handling them. Actions are split into three streams: the pose of the end effector, the joints of the end effector itself, such as a gripper or a dexterous hand, and the joints of the lower body. Each stream is discretised by its own residual vector quantisation model, and the resulting token sequences are aligned in time and fed to the model together. The card presents this as the link between understanding a scene, predicting its changes and coordinating the whole body. The model card shows demonstrations of cleaning a table, folding a towel, packaging a phone and placing plates. UnifoLM-ER-Flow is part of Unitree's UnifoLM-WLA-1.0 collection on Hugging Face.

Unitree Robotics (Hugging Face) (original source)OFFICIAL
OPEN SOURCE

Unitree publishes UnifoLM-ER-1, a 4B vision-language model for spatial reasoning in embodied settings

Unitree Robotics has published the weights of UnifoLM-ER-1 on Hugging Face. The model card describes a 4B-parameter vision-language model built on Qwen3-VL-4B, distributed as BF16 Safetensors under the Apache 2.0 licence. The repository was created on 11 September 2026 and belongs to Unitree's UnifoLM-WLA-1.0 collection. According to Unitree, the model saw more than 5 million training samples covering point prediction in images, object detection, reasoning across several images, 2D trajectory prediction, 3D object detection and spatial question answering over multiple images. Those embodied tasks were mixed with general image-text data during training, which Unitree says keeps the model's general vision-language ability while strengthening spatial understanding and reasoning for robots. The card reports scores on 16 perception and understanding benchmarks, and Unitree says the model leads open-source models on seven of them. Set against its own base, Qwen3-VL-4B, the table shows RoboVQA rising from 47.7 to 62.4, Where2Place from 63.0 to 82.0 and ERQA from 41.3 to 50.0. The base model still scores higher on VSI-Bench, 59.3 against 54.2, and on RealWorldQA, MME and MMMU_VAL. The footnotes set out how the comparison was assembled: figures for other open models come from their technical reports or papers, proprietary models were tested through their official APIs, and BLINK scores are averaged over the Relative Depth and Spatial Relation subtasks only. No inference provider serves the model yet; one Hugging Face Space uses it for a demo.

Unitree Robotics (Hugging Face) (original source)OFFICIAL
OPEN SOURCE

AGIBOT Releases GE

--> AGIBOT Releases GE-AGIBOT Innovation (Shanghai) Technology Co., Ltd. Home page Products AGIBOT A AGIBOT X AGIBOT Genie AGIBOT D1 AGIBOT C5 AGIBOT Accessories AGIBOT Premium Selection Document Center Security Center 智元远征A2 Max 2 AGIBOT A3 Silicon-based star born for the stage AGIBOT A2 Ultra Full-Size Humanoid Robot AGIBOT A2 Lite Master of Performance AGIBOT A2-W Flexible Manufacturing Robot AGIBOT X1 Full-Stack Open-Source Robot AGIBOT X2 series Fully Intelligent & Agile Robot AGIBOT G2

AGIBOT A2-WAGIBOT A3AGIBOT Genie G2AGIBOT X2AGIBOT (original source)OFFICIAL
OPEN SOURCE

Unitree open-sources UnifoLM-WLA and releases the UnifoLM-ER models

Unitree published the UnifoLM-WLA repository on GitHub on 11 September 2026, with the weights of two models, UnifoLM-ER-1 and UnifoLM-ER-Flow, on Hugging Face the same day; the model modules and code for training action experts followed on 20 September. Unitree describes UnifoLM-WLA-1.0 as a 6-billion-parameter humanoid foundation model, trained on about 2,500 hours of real-robot data, that coordinates 64 tasks across tabletop and whole-body manipulation with two-finger grippers and five-finger hands. UnifoLM-ER-1, built on Qwen3-VL-4B, targets spatial understanding and reasoning in embodied settings. The code is released under the Apache License 2.0.

ManipulationImitation LearningDexterous HandsGitHub (original source)OFFICIAL

1 UPDATE

RESEARCH

CAP keeps humanoid locomotion working when depth sensing fails (CoRL 2026)

CAP is a single locomotion policy for humanoids that keeps crossing complex terrain when its depth input is partly or intermittently corrupted. A learned denoiser reconstructs clean depth from the corrupted signal, a proprioceptive encoder supplies body-state information that does not depend on depth, and training exposes the policy to the full range of perception quality. The authors report that it degrades more smoothly than a switching baseline in simulation and demonstrate it on a Unitree G1 indoors and outdoors. The paper is accepted at the Conference on Robot Learning 2026.

Unitree G1LocomotionSim-to-realComputer VisionarXiv (original source)REVIEWED

1 UPDATE

1 UPDATE

PRODUCT

XPENG commissions humanoid production lines in Guangzhou; first IRON walks off the line

XPENG announced on 8 September 2026 that its humanoid robot production lines in Guangzhou had been commissioned and that the first IRON robot had walked off the line on its own. The company says more than 80% of core processes are automated, that mass production is scheduled by the end of 2026 with initial use in its own stores and campuses, and that market launch and delivery in China and overseas are planned for 2027.

IRON (next-generation)XPENG (XPeng Inc.) - robotics business (original source)OFFICIAL

1 UPDATE

RELEASE

unitree_sdk2_rs 0.3.0 adds Rust RPC clients for eight Unitree robots

Unitree's Rust bindings for its SDK2, unitree_sdk2_rs, reached version 0.3.0 on 4 September 2026. The release generates message types for the unitree_api package, adds RPC clients for eight models including the G1, H1, H2 and R1 humanoids, and changes the licence to BSD-3-Clause.

Unitree R1Unitree H2Unitree H1Unitree G1GitHub (original source)OFFICIAL

1 UPDATE

PRODUCT

Galbot opens pre-orders for ET1 interactive humanoid

Galbot opened pre-orders for ET1, a 1.25 m, 32 kg interaction humanoid with 26 DoF (plus 12 with dexterous hands) powered by its AstraBrain-Agent and AstraBrain-WBC models.

Galbot ET1Whole-body ControlGalbot (original source)OFFICIAL

HISTORICAL RECORD

Earlier developments

Developments WBH recorded after the fact, dated by their original publication. They are part of the timeline, not current news.

1 UPDATE

OPEN SOURCE

Figure launches Index, a paid crowd-sourced human-activity video dataset app

Figure made its data-collection app public as Index on Google Play and the App Store. Figure says 'Creators' have uploaded over 16M videos, been paid $15M, and that it will spend over $1B in the next 12 months on data and compute. The dataset is Figure-exclusive, not a public release.

Imitation LearningFigure AI (original source)OFFICIAL

1 UPDATE

PRODUCT

1X details 25-DoF tendon-driven hands that will ship on every NEO

1X published an engineering write-up of NEO's new hands: 25 DoF (22 in fingers and palm plus 3 at the wrist), quasi-direct-drive tendons at roughly 5:1-15:1 gear ratios, fully backdrivable and force-controlled, with tactile and shear sensing. 1X says these hands will ship on every NEO.

NEOActuatorsDexterous HandsManipulation1X Technologies (original source)OFFICIAL

2 UPDATES

PRODUCT

Apptronik unveils Apollo 2 and opens expanded Robot Park data facility in Austin

Apptronik unveiled Apollo 2 in bipedal and wheeled-base configurations and opened a ~90,000 sq ft Robot Park in Austin, where Apollo 2 fleets collect teleoperated and autonomous data for Google DeepMind's Gemini Robotics models. It named Apollo 3 as its next-generation commercial product.

Apollo 2TeleoperationApptronik (original source)OFFICIAL
INDUSTRY

Figure 03 begins parts-sequencing work at BMW Plant Spartanburg

Figure says Figure 03 arrived in Hall 52 at BMW Group Plant Spartanburg, moving from Figure 02's sheet-metal loading to a parts-sequencing use case that includes pulling a heavy cart, run by Helix 02.

Figure 03ManipulationWhole-body ControlFigure AI (original source)OFFICIAL

1 UPDATE

RELEASE

AGIBOT security advisory SA-2026-0001: unauthenticated MQTT on X2 Ultra v0.9.7, fixed in v1.0.0

AGIBOT's security centre lists advisory SA-2026-0001: the MQTT service on AgiBot X2 Ultra v0.9.7 allowed unauthenticated subscribe/publish (medium severity), fixed in v1.0.0; mitigation is disabling unauthenticated access on port 1883.

AGIBOT X2SafetyAGIBOT (original source)OFFICIAL

1 UPDATE

INDUSTRY

Figure signs commercial agreement with Catalyst Brands for distribution-centre humanoids

Figure announced a commercial agreement to deploy humanoids in Catalyst Brands' logistics network, starting in Reno, Nevada. Brookfield is an investor in both companies.

Figure AI (original source)OFFICIAL

1 UPDATE

PRODUCT

EngineAI rolls out first production batch of T800 humanoids

EngineAI reports that the first batch of T800 full-size humanoids rolled off its production line, framed as a step toward 10,000-unit-scale delivery. The T800 was introduced at CES 2026.

EngineAI T800EngineAI (original source)OFFICIAL

1 UPDATE

INDUSTRY

Humanoid signs binding deal to deploy wheeled HMND 01 robots at Schaeffler from December 2026

Humanoid signed a binding phased deployment and supply agreement with Schaeffler. The first phase runs December 2026 to June 2027 at two German sites, with a target of a four-digit number of wheeled units by 2032; Schaeffler becomes a preferred actuator supplier.

HMND 01 Alpha WheeledHumanoid (original source)OFFICIAL

1 UPDATE

PRODUCT

1X starts full-scale NEO production at its Hayward, California factory

1X said its NEO Factory in Hayward, California has begun full-scale production, with initial robots going to its R&D lab and internal home testing and first customer shipments planned for 2026. The company claims capacity for 10,000 NEOs per year with plans for more than 100,000 by the end of 2027, and says it sold out its first-year capacity of over 10,000 units within five days of the October 2025 launch. Early Access costs $20,000, with a $499/month subscription option.

NEO1X Technologies (original source)OFFICIAL

1 UPDATE

PRODUCT

Figure says BotQ has delivered 350+ Figure 03 units and reached one robot per hour

Figure reports its BotQ facility delivered over 350 Figure 03 robots and went from one robot per day to one per hour in under 120 days, with end-of-line first-pass yield above 80%. The same post adds perception-conditioned whole-body control (S0 using stereo from head cameras) trained in simulation.

Figure 03LocomotionReinforcement LearningSim-to-realFigure AI (original source)OFFICIAL

1 UPDATE

PRODUCT

AGIBOT launches A3 full-size humanoid at 2026 Partner Conference

At its 2026 Partner Conference AGIBOT introduced the A3 humanoid (173 cm, 55 kg, 10 h endurance, 10-second battery swap) alongside G2 Air, OmniHand 3 Ultra-T, the D2 Max quadruped, the MEgo data-collection system and eight foundation models.

AGIBOT A3AGIBOT (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

AGIBOT unveils GO-2 embodied foundation model

AGIBOT introduced Genie Operator-2 (GO-2), a successor to GO-1 that unifies reasoning and action in one architecture, trained on tens of thousands of hours of interaction data; AGIBOT says core parts were accepted at CVPR 2026 and ACL 2026.

AGIBOT (original source)OFFICIAL

1 UPDATE

OPEN SOURCE

AGIBOT open-sources AGIBOT WORLD 2026 dataset (themes released April-September 2026)

AGIBOT released AGIBOT WORLD 2026, a heterogeneous real-world robot dataset using free-form teleoperated collection in commercial, home and everyday scenes, on Hugging Face. Theme 2 and Theme 3 (real-world RL, 2026-09-03) followed.

Imitation LearningAGIBOT (original source)OFFICIAL

1 UPDATE

PRODUCT

UBTECH reports 1,079 full-size humanoids sold in 2025

UBTECH's 2025 annual results announcement on HKEX reports roughly RMB 820 million revenue from full-size embodied humanoid products and services and 1,079 units sold in 2025.

Walker S2UBTECH Robotics (HKEX filing) (original source)OFFICIAL

1 UPDATE

PRODUCT

Noetix delivers first Bumi humanoids to consumers at AWE 2026, deepens JD.com partnership

At AWE 2026 in Shanghai, Noetix founder Jiang Zheyuan handed over Bumi units to customers, and Noetix and JD.com announced deeper cooperation on consumer humanoid sales.

Noetix N2Noetix BumiNoetix Robotics (original source)OFFICIAL

1 UPDATE

1 UPDATE

1 UPDATE

OPEN SOURCE

1X makes its video-pretrained World Model NEO's policy for unseen tasks

1X described 1XWM, a text-conditioned video world model integrated into NEO as a robot policy, which it says generalises to novel objects and tasks without large-scale robot pretraining data.

NEOImitation Learning1X Technologies (original source)OFFICIAL

1 UPDATE

PRODUCT

NEURA unveils Porsche-designed 4NE1 humanoid and 4NE1 Mini at CES 2026, opens reservations

NEURA Robotics showed the next generation of 4NE1, designed with Studio F.A. Porsche, and a compact 4NE1 Mini at CES 2026. Reservation pages list 98,000 € (1-19 units) for 4NE1 Gen 3.5 with availability expected end of 2026, and 19,999 €/29,999 € for 4NE1 Mini Standard/Pro.

4NE1 Gen 3.54NE1 MiniIsaac SimROS 2NEURA Robotics (original source)OFFICIAL

1 UPDATE

PRODUCT

XPENG unveils next-generation IRON humanoid with 82 DoF, targets mass production by end-2026

At its 2025 AI Day in Guangzhou, XPENG presented the next-generation IRON humanoid with a humanoid spine, bionic muscles and flexible skin, 82 degrees of freedom, 22-DoF hands, three Turing AI chips and an all-solid-state battery. XPENG said it aims for large-scale mass production of advanced humanoids by the end of 2026.

IRON (next-generation)ActuatorsDexterous HandsXPENG (XPeng Inc.) - robotics business (original source)OFFICIAL

2 UPDATES

RELEASE

Isaac Lab 2.3 adds a G1 loco-manipulation environment

Isaac Lab 2.3.0, released on 28 October 2025 for Isaac Sim 5.1, added a Unitree G1 environment that combines reinforcement-learning locomotion with inverse-kinematics manipulation for the Mimic demonstration workflow, and teleoperation support for the G1 with Inspire five-finger hands.

Unitree G1Isaac SimManipulationImitation LearningTeleoperationGitHub (original source)OFFICIAL
PRODUCT

1X opens NEO pre-orders at $20,000 or $499/month for US homes

1X launched NEO for pre-order with a $20,000 Early Access purchase (priority 2026 delivery) or $499/month subscription, shipping first in the US in 2026 and other markets from 2027. 1X later said it booked a year of capacity (10,000 units) in five days.

NEO1X Technologies (original source)OFFICIAL

1 UPDATE

PRODUCT

Figure unveils Figure 03, its third-generation humanoid

Figure introduced Figure 03 with a redesigned camera system, palm cameras, in-house fingertip tactile sensors, soft washable covers, 2 kW wireless inductive charging through the feet and a design for high-volume manufacturing at BotQ.

Figure 03Dexterous HandsSensorsFigure AI (original source)OFFICIAL

1 UPDATE

RELEASE

unitree_ros2 v0.3.0 adds G1 examples and aligns hand messages with unitree_sdk2

Released on 15 August 2025, unitree_ros2 v0.3.0 added G1 examples for audio, dual-arm control, ankle swing, locomotion and arm actions, plus a Dex3 hand example, and changed the hand command, hand state and pressure-sensor messages to match unitree_sdk2. It also fixed a Foxy-specific bug in topic statistics.

Unitree G1ROS 2Dexterous HandsGitHub (original source)OFFICIAL