RoboSignal Daily · 2026-11-10 · Tuesday
ROBOSIGNAL
Apple Research Team Discloses Progress on Robot 'Brain' System
On November 4, Apple's research team submitted a preprint titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system for humanoid robots. The system was validated on Unitree G1 robots, enabling actions learned in simulation to be executed in unfamiliar real-world environments. Industry analysts note that such 'embodied brains' are central to advancing embodied intelligence from demonstration to practical application.
Robot Models
LeRobot 0.7: Interactive Rollouts for Mid-Run Steering
LeRobot 0.7 releases interactive rollouts allowing users to steer robot policies mid-run without reset. Features include /subtask switching, /vqa visual queries, /autosteer goal planning, and episode recording. Any VLM (transformers, vLLM, or inference providers) can serve as a planner. FineART-VLA, trained on 1,718 hours of data with 533,913 subtask labels from scale_AI and Hugging Face, verbalizes next subtasks before execution.
Open source multi-floor SLAM for ROS 2 released by EasyNav team
The Open Robotics account relays a release announcement from the EasyNav team regarding StrataSLAM, an open-source multi-floor SLAM solution for ROS 2 licensed under Apache-2.0.
Research
Apple Research Team Discloses Progress on Robot 'Brain' System
On November 4, Apple's research team submitted a preprint titled 'Perceptive Humanoid Motion from Simulation' to arXiv, detailing the HALO autonomous control system for humanoid robots. The system was validated on Unitree G1 robots, enabling actions learned in simulation to be executed in unfamiliar real-world environments. Industry analysts note that such 'embodied brains' are central to advancing embodied intelligence from demonstration to practical application.
IEEE Spectrum — RoboticsCan Agentic Robotics Solve Robotics' Data Scaling Problem?
At the 2026 Conference on Robot Learning, UC Berkeley and NVIDIA announced agentic robotics, a method using large language models to write robot control code directly from natural language prompts. This approach aims to overcome the plateau of imitation learning by treating robot programming as a coding problem rather than a data collection one. The system decomposes tasks into sub-tasks handled by specialized coding agents (e.g., perception, manipulation) and iteratively refines code in simulation, even adjusting simulation parameters to match real-world video. While testing showed up to 99 percent success rates matching human-engineered performance on specific tasks, the authors acknowledge challenges with sim-to-real physics gaps and AI-generated code reliability.
ETH Zurich — Robotics research newsFirst-partyDissolving blood clots with flow-generating microbots
Researchers at ETH Zurich have developed magnetically guided microbots made of a biocompatible gel containing iron oxide nanoparticles.
UC Berkeley's SuperMimic Reduces Whole-Body Control Training Time
Researchers at UC Berkeley have introduced SuperMimic, a full-body control architecture based on a single differentiable world model. The method aims to replace compute-intensive model-free reinforcement learning by learning shared dynamics across diverse human movements. In benchmarks involving over 129,000 human motions, the system reportedly achieved superior performance compared to prior methods like NVIDIA's SONIC while reducing training from multi-GPU clusters running for days to a single GPU in approximately three hours.
arXiv Robotics — research abstractsCS-Pi: Constraint-Satisfying Flow Policy Sampling
This paper introduces CS-Pi, a novel sampling technique for flow-matching robot policies that enforces safety constraints at inference time without retraining. The method projects actions to the nearest safe state and adds noise scaled by the projection distance before resampling, balancing efficiency with action plausibility. Experiments on a 7-DoF shelf pick-and-place task and zero-shot Franka hardware deployment show CS-Pi achieves 81% and 55% safe success rates respectively, outperforming standard projection (55% and 40%) while maintaining plausible action distributions.
Simulation & World Models
Skild AI open-sources Steel, a physics solver built with NVIDIA
Skild AI has released Steel, an internal physics solver developed with NVIDIA. The tool claims to run simulations up to 13x faster than PhysX while using about 3x less GPU memory, aiming to enable scalable simulation for manipulation tasks.
Deployments
Over 40 Jingchu Robots Delivered Concentrately; To Deploy in Multiple Locations Inside and Outside the Province
On November 3, Hubei Jingchu Humanoid Robotics Co., Ltd. completed the delivery of over 40 robots comprising three models。
Industry
CNPC-affiliated entities hold 60% stake in new joint venture with Unitree and Beijing Fanjie for oil and gas embodied AI robots
Beijing market regulator recently published a notice on a proposed joint venture between China Petroleum Group Kunlun Manufacturing Co., Ltd. (with Kunlun Capital as a consistent actor) and Unitree Technology and Beijing Fanjie Intelligent Technology Co., Ltd. The new entity will focus on industrial embodied intelligent robot solutions, with Kunlun-affiliated parties holding 60% equity. The arrangement aims to integrate hardware, AI models, and site acceptance standards within one corporate structure to develop replicable products for oil and gas operations.
Korea Advances Physical AI Across Industries With NVIDIA
NVIDIA is collaborating with major Korean industry leaders to integrate accelerated computing, AI physics, and simulation into real-world production. Samsung Electronics is using these tools for semiconductor engineering and materials research, achieving significant speedups in thermal analysis and quantum chemical calculations. Kakao Mobility is developing a level 4 autonomous driving system by combining its proprietary software with NVIDIA's vehicle computing platform and safety technologies. LG Electronics has demonstrated robotic manipulation tasks using the Isaac GR00T model on its CLOiD robot within an actual manufacturing environment, establishing a data flywheel for continuous robot learning and validation.
Humanoid robots gathering dust in warehouses as commercial prospects cool down
Despite a spectacular debut at the Spring Festival Gala and Unitree Technology's successful transition to humanoid manufacturing with over 5,500 units shipped in 2025, the commercial market for humanoid robots is facing a severe downturn. While Unitree achieved significant revenue growth and profitability in 2025 before its stock market listing, the broader industry is experiencing a shift where customer demand no longer matches the supply of available robots, leading to high idle rates.
Briefs
- What Limits mmWave Human-State Sensing for Robot Perception?arXiv Robotics — research abstracts
- Can't be inaccurate: how 2 tiny robot arms could halve digestive cancer surgery timeSurgical robotics — News discovery
- Autonomous electric hauling company Humble Robotics raises $51M Series ARobotics 24/7 — Industry reporting
- Imo University Establishes Africa's First Robotic Surgery AcademySurgical robotics — News discovery
- NVIDIA Hosts AI Day in Seoul, Announces Expanded Partnerships with South Korean Government and Major CorporationsIT Home — Robotics
- Precision Spatial Tracking for Autonomous Systems: Training Humanoid Robots and Drones with Optical Motion CaptureRobotics & Automation News
- Samsung, LG seek robotics talent overseas as commercialization push growsRobotics — Web news discovery
- EDGE Sound Research raises $2 million to expand physical AI perception technologyRobotics & Automation News
- It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for AssemblyarXiv Robotics — research abstracts
- ANYbotics begins deliveries of ANYmal X robot for hazardous industrial environmentsRobotics & Automation News