RoboSignal Daily · 2026-11-12 · Thursday
ROBOSIGNAL
Boston Dynamics equips Atlas humanoid robot with wheels, enabling it to jump and spin
On November 9, 2026, Boston Dynamics released a video demonstrating an Atlas humanoid robot modified with wheels attached to its ankle actuators. Engineers used custom adapters to install standard electric scooter wheels without major hardware changes or software rebuilding, only swapping feet for wheels in simulation. The modified Atlas can perform aerial rotations and lateral sprints by controlling wheel steering instead of dragging feet. Actions are derived from motion capture data, including foot sliding motions adapted for rolling movement. A trajectory tracking controller enables the robot to jump and spin successfully on its first attempt with recorded human motion data. This platform is intended for research rather than commercialization.
Robot Models
Ant Group's Lingbo Opens-Source Tactile Foundation Model LingBot-Touch for Robot Dexterous Manipulation
On November 10, Ant Group's Lingbo officially open-sourced the tactile foundation model LingBot-Touch to address gaps in robot tactile perception. Pretrained on nearly 5,000 hours of real-world tactile data, the model standardizes signals from magnetic, visual-tactile, and piezoresistive sensors into a unified representation. Tests on Sharpa Wave and XHand robots across 26 tasks showed that integrating LingBot-Touch improved average task success rates by 14–26 percentage points compared to non-tactile approaches.
Giving instructions to a robot mid-task: LeRobot 0.7
On November 5, the LeRobot team at Hugging Face released version 0.7.0 of their robot learning and motion testing software.
Research
arXiv Robotics — research abstractsCAFE: Closed-Loop Autonomous Task Formulation and Execution without Instance-Specific Instructions
This research introduces CAFE, a system for scene-driven robot autonomy that selects and executes tasks within predefined categories like cleaning and storage without instance-specific goals. The architecture maintains online memory of objects and zones, using a rule-guided Trigger-LLM to admit formulation from evidence while Draft, Complete, and Audit modules propose goals, resolve details, and check dependencies. Evaluation on 12 AI2-THOR scenes compares three backbones and task-complexity tiers using coverage, constraint, and execution metrics, with results showing Tri-LLM achieving higher GCR scores than baselines. The paper includes controls on completion, auditing, zones, feedback, triggering, and prompt examples, alongside LimoCobot traces illustrating storage, cleanup, and recovery from failed grasps.
Data & Training
arXiv Robotics — research abstractsEmbodiedIntentBench: Benchmarking Intent-Driven Runtime Adaptation for Mobile Robotic Missions
This research presents EmbodiedIntentBench, a benchmark designed to evaluate intent-driven runtime adaptation strategies in mobile robotic missions.
Robots & Hardware
Boston Dynamics equips Atlas humanoid robot with wheels, enabling it to jump and spin
On November 9, 2026, Boston Dynamics released a video demonstrating an Atlas humanoid robot modified with wheels attached to its ankle actuators. Engineers used custom adapters to install standard electric scooter wheels without major hardware changes or software rebuilding, only swapping feet for wheels in simulation. The modified Atlas can perform aerial rotations and lateral sprints by controlling wheel steering instead of dragging feet. Actions are derived from motion capture data, including foot sliding motions adapted for rolling movement. A trajectory tracking controller enables the robot to jump and spin successfully on its first attempt with recorded human motion data. This platform is intended for research rather than commercialization.
Unitree G1 Robot Withdraws from Mount Tai Race Due to Overheating
At the inaugural World Robot Mount Tai Competition in Taian, Shandong, a Unitree G1 development version robot withdrew due to overheating during ascent. Experts attribute this to heat accumulation from high-load joint motors on uneven stone steps rather than ambient temperature, noting that compact designs prioritize agility over sustained thermal dissipation. The incident underscores gaps in autonomous adaptation to irregular terrain, error correction without human intervention, and the engineering trade-off between lightweight actuation and continuous work endurance. Industry observers suggest that practical value will be measured by stable effective working hours, task frequency, and recovery from anomalies rather than peak performance metrics like speed or jump height.
IT Home — RoboticsLoongson Unveils Domestic Computing Platform for Intelligent Robot Vehicle with Depth Vision and Object Grasping
On November 11, Loongson announced that its self-developed intelligent robot vehicle has completed a full-stack port of ROS2 Jazzy. The system integrates mapping, localization, joystick remote control, robotic arm operation, and depth vision, all orchestrated by the Loongson 2K3000 processor. This domestic computing solution validates a reusable development approach for industrial automation and smart warehousing prototypes.
Jianzhi Robotics Launches DAS 2.0 System to Address Data Scarcity in Embodied AI
Jianzhi Robotics has introduced the DAS 2.0 system, comprising the Dex 2.0 dexterous glove and Ego 2.0 first-person view head-mounted unit, designed to capture aligned physical causal chains for embodied AI training. The system addresses data scarcity by ensuring complete observation-action-contact-feedback loops through direct kinematic measurement and reliable tactile sensing, aiming to provide high-fidelity human skill translation for robot learning.
Deployments
Warehouse Logistics as the First Commercial Test for Embodied AI
The article identifies warehouse logistics as a leading commercial application scenario for embodied AI, citing policy support and demonstrations by companies such as Figure, Xingdong Jiyuan, and JD Logistics. It explains why semi-structured environments suit current robot capabilities, outlines the involvement of three player types—AI startups, automation firms, and logistics operators—and discusses challenges in integrating robots with existing WMS/WCS systems for scalable deployment.
Industry
WuJing Dynamics Halts Operations Amid Wage Arrears and Failed Deliveries, Highlighting Embodied AI Bubble
WuJing Dynamics, a robotics company founded in February 2026, has ceased operations after failing to deliver on promised products and paying employees since August. Despite earlier announcements of nearly 1 billion yuan in orders and 110 million yuan in funding, the firm faces severe cash flow issues, including unpaid wages totaling approximately 8 million yuan for dozens of staff. This case reflects a broader trend in the embodied AI sector where inflated framework agreements and capital market hype mask commercialization shortcomings, leading to liquidity crises among peers.
Foreign Tier 1 Suppliers Begin 'Selling Scenarios' to Robotics Companies
At the 2026 China International Import Expo, automotive Tier 1 suppliers like Aptiv, Schaeffler, Bosch, and Valeo are transitioning from selling components to offering production scenarios as a resource for robotics companies. Aptiv has deployed Tars A-series robots in its actual lines for connector assembly, while Schaeffler opened a new facility combining manufacturing with robot training. Bosch and Valeo are using existing factories for trials and data collection. The article highlights that while embodied models define capability limits, hardware reliability, thermal management, and long-term stability in real production environments are the decisive factors for industrial adoption. Suppliers now treat their production lines as mandatory testbeds, requiring robots to meet strict quality, efficiency, and safety standards before full deployment.
IT Home — RoboticsBeijing Issues Action Plan to Cultivate Embodied AI Robot Industry Through Dedicated Listing Support
On November 11, Beijing's Economic and Information Bureau issued the 'Action Plan for High-Quality Development of the Embodied Intelligent Robot Industry (2026-2028)'. The plan establishes a one-on-one service mechanism to support companies in listing and aims to cultivate over 200 technology-driven SMEs in key areas including complete machines, critical components, embodied AI models, and data collection. It also promotes technological integration across routes such as visual-language-action (VLA) models, world models, and reinforcement learning.
Briefs
- Multi-Objective Adaptation After Deployment: A Position on World Models for Remote Robot TeamsarXiv Robotics — research abstracts
- Grip, Trace, Write: Force-Aware Imitation and Reinforcement Learning for High-DoF Robotic HandwritingarXiv Robotics — research abstracts
- ASPIRE: An Agentic System-Identification Platform for Iterative Reasoning and ExplorationarXiv Robotics — research abstracts
- How Much of Your VLA Is kNNarXiv Robotics — research abstracts
- Capturability-Aware Footstep Planning in Cluttered Environments: A Recursively Feasible Sequential MPC on Axis-Aligned Safe CorridorsarXiv Robotics — research abstracts
- Look Only Where You Plan for Real-Time End-to-End Urban DrivingarXiv Robotics — research abstracts
- Efficient Learning of Differentiable MPC Policies on the GPU via Smoothed ADMM and First-Order Reinforcement LearningarXiv Robotics — research abstracts
- Tactile Sensors Are an Effective Substitute for Wrist Cameras in Contact-Rich or Occluded ManipulationarXiv Robotics — research abstracts
- Maze Runner: Autonomous Maze Navigation with Line-Based Localization and Control Barrier FunctionsarXiv Robotics — research abstracts
- OfficeTwin: Realizing 3D Reconstruction of Redundant Scenes through Partial Viewpoint PerceptionarXiv Robotics — research abstracts