Research & Signals31 articles

Berkeley and NVIDIA release AI agents that code industrial robots in simulation
UC Berkeley and NVIDIA have released Graph-as-Policy (GaP) and VABAR, an open-source framework that enables AI coding agents to program industrial robots by iterating in simulation. The approach eliminates the need for large training datasets and produces human-readable code that engineers can audi…

Boston Dynamics tests wheeled Atlas to merge classical control with whole-body RL
Boston Dynamics released video of its Atlas humanoid using scooter wheels attached to ankle actuators, controlled via a hybrid system that merges classical model-based control with whole-body reinforcement learning. The experiment demonstrates that Atlas can dynamically adjust pose and force in rea…

AGIBOT's GO-3 Preview tops three embodied AI benchmarks with full-lifecycle learning

Robot moth grows UV sensors using circulating artificial blood

LiberAI launches Liber-0 world action model trained on 2.2M hours of human experience

Leiden study confirms autonomous UV-C robot meets hospital disinfection standards

ETH Zurich trains robotic hand to crawl using its own fingers

Skild AI uses self-play to train robots beyond human capability

Tesla patents scalable touch-sensitive skin for robot hands

MIT's SANDO drone planner guarantees collision avoidance in unknown terrain

Swiss researchers develop flexible fiber motor for soft robotics

MIT robot swims using living muscle cells controlled by light

Cyborg cockroaches deliver drug injections in search-and-rescue tests

Autonomous vehicles learn to take passenger requests via LLM interface