Researchers at the University of Illinois Urbana-Champaign's Grainger College combined reinforcement learning with control theory to guarantee stability for wheeled, flying and six-legged robots. The method improved tracking performance by up to 15% over existing approaches while keeping robots within reasonable operating bounds.
Nvidia has expanded its open model portfolio across vision, language, robotics, biomedicine and quantum computing, aiming to help European enterprises and researchers build and deploy custom applications with local data.
Chipsea Technology says its EC chips can be extended to external interfaces and sensor control in humanoid robots, though the application is still in early commercialization and has not generated scaled revenue.
Mech-Mind Robotics is applying its embodied AI "Eye-Brain-Hand" system to inline glue bead inspection in automotive sheet metal production. Glue bead defects can compromise body sealing and structural integrity, while traditional offline sampling delays detection and fixed-view cameras struggle with corners and curved surfaces. The system inspects in real time, synchronized with the dispensing process, at speeds of up to 1,500 mm/s, catching defects that sampling and fixed cameras may miss.

ACE Robotics, founded by SenseTime co-founder Xiaogang Wang, has released ACE-Brain-0.6, an embodied AI model built by supervised fine-tuning Qwen3.5-9B for object identification, image localization and robot interaction-point prediction. The company says the model, trained on roughly 1 million spatial-grounding samples, beat GPT-6 Astra on six of eight grounding benchmarks and scored 74.53 and 94.13 on two evaluation sets.
Samsung Electronics is developing a humanoid robot capable of 24/7 operation and is testing it at its Seoul R&D center, with a possible debut at CES 2027 as the company aims to catch up in the robotics market.
Agility co-founder Jonathan Hurst explains in IEEE Spectrum why the company's latest humanoid, Digit 5, is designed without running capability, even though his earlier robots could run. The article walks through the redesign decisions behind Digit 5, framing the trade-offs the team made between dynamic locomotion and the practical requirements of a commercial humanoid robot.

Daimon Robotics is building a pipeline that connects tactile sensing, data collection, simulation and AI models to provide infrastructure for tactile intelligence. The company aims to turn physical interactions into training data for more reliable manipulation, targeting assembly, packaging and other contact-rich tasks while collaborating with downstream developers.
Researchers from AgiBot, NVIDIA, Princeton, UT Austin, and the University of Toronto have released WorldOracle, an action-driven benchmark that evaluates world models by comparing predicted videos with real robot execution. The platform is open for submissions and includes 335 real operation trajectories, about 40% of which contain failures or suboptimal actions.
Daxiao Robotics has released ACE-Brain-0.6, a 9B embodied foundation model trained on about one million verifiable spatial-anchoring samples. It beat OpenAI's GPT-6 Astra on six of eight grounding evaluations and averaged 74.53 on seven public benchmarks versus 73.68.
Daxiao Robotics has released ACE-Brain-0.6, an embodied base model built on Qwen3.5-9B and trained with about one million spatial-anchoring samples. The company says it outperforms OpenAI's GPT-6 Astra on six of eight grounding benchmarks.
Daxiao Robotics has released ACE-Brain-0.6, an embodied foundation model fine-tuned from Qwen3.5-9B with about one million spatial-anchoring samples. The company says it outperforms OpenAI's GPT-6 Astra on six of eight grounding benchmarks.
D-Robotics has published the full recording of the first session of D-Robotics Frontier, which drew attendees interested in its Uranus platform. The talk explores the company's approach to interactive, data-driven robot simulation and the challenges of building world models for robotics. The recording is available on YouTube.