Agility Robotics states that every Digit humanoid runs as its own secured system, with a unique cryptographic identity baked into the hardware so that no two units share an identity. CTO Pras Velagapudi explains that the design keeps a security incident contained to one robot instead of letting it spread across the whole fleet.

Skild AI has released Steel 0.2, the GPU-accelerated rigid-body solver inside the open-source Newton Physics Engine used for robot learning in NVIDIA Isaac Lab. The update ships 81 simulation environment examples, interactive documentation and migration instructions for coding agents, plus magnetic contacts that latch, hold and release, and closed-loop joints for grippers and parallel robots.

Skild AI says its Steel tool can be used to create simulation environments or convert existing PhysX and MuJoCo environments. Steel is a GPU-accelerated rigid-body solver in the open-source Newton Physics Engine, aimed at robot learning in NVIDIA Isaac Lab from @NVIDIARobotics.
Academician Guo Song, Estun, and Feidu Technology discussed in a joint interview how physical AI can bring embodied intelligence from simulation to factory floors, stressing data loops, sim-to-real transfer, and industrial deployment.
STEP (002527.SZ) has invested in machine vision, adaptive robot control, and autonomous trajectory planning, but these have not yet significantly impacted its performance.
Zero Power has open-sourced a 'physical feel' technology for embodied AI, aiming to give robots more intuitive touch and force sensing while reshaping the competitive landscape.
Apptronik has built a robot park training ground where multiple Apollo humanoid robots repeatedly perform tasks in simulated industrial environments to continuously collect real-world data. The company has raised over $935 million at a valuation of about $5.5 billion, with customers including Mercedes-Benz and Jabil, as real-world data becomes a key bottleneck for scaling humanoid robots from pilots to factories.
Engineers at UC San Diego have built the fastest robot made entirely of soft materials, reaching 4.72 meters per second. The six-legged, 3D-printed machine runs on sand, mulch and concrete, swims and walks underwater, and operates without electronics.
A University of Potsdam research team argues that acceptance of humanoid robots depends on how people experience interaction with them, not only on technical performance. The study says acceptance emerges from a dynamic interplay of perception and social interaction over time.
A Hong Kong University of Science and Technology team reports a port-specific industrial robot that matched human drivers over 60,000 real operations, with the work featured on the latest Science Robotics cover.
A humanoid robot at a Shenzhen park refreshment stand took a cup and pulled a lever to shape an ice cream spiral before handing it to a customer. The demonstration reflects how robotics firms in China’s tech hub are training machines to learn from human actions.