GRAB

Pollen Robotics Releases Grabette to Lower Barriers for Open Robot Manipulation Datasets

Pollen Robotics launched Grabette, a handheld open hardware system that records 6-DoF manipulation demonstrations without a robot, directly feeding LeRobot datasets on the Hugging Face Hub. The device pairs a fisheye camera and RGBD sensor with IMU tracking at roughly 490€ BOM cost, extending the UMI approach into a browser-based pipeline. By releasing both Grabette and its motorized twin Gripette, the team targets the data bottleneck that currently limits scalable visuomotor policy training acr

AGIBOT Fields A3 Ultra Humanoid and G2 Max Task Platform at WAIC to Push Repeatable Industrial Workflows

AGIBOT launched four embodied AI products at WAIC 2026, headlined by the A3 Ultra full-size humanoid and G2 Max heavy-payload task robot, to shift industry focus from single demos to manufacturable, integrable systems. The Shanghai company, which has now built 15,000 units since its 2023 founding, pairs its Three Intelligences in One architecture with NVIDIA Thor processors to target commercial service and industrial material-handling deployments.

LeRobot v0.6.0: Leaner Installs and Refactored RL Stack

This update makes LeRobot leaner and much easier to tinker with, as the default installation no longer bundles heavy training dependencies. The biggest win is the fully refactored reinforcement learning stack, which is now built like swappable Lego blocks so you can swap out a single network without breaking the rest. Plus, the default GR00T embodied AI model gets a smarter N1.7 upgrade with better action generalization. • Leaner installs: pip install now only grabs core features; training and dataset extras must be explicitly requested to avoid bloat and conflicts. • Modular RL stack: The reinforcement learning code is fully decoupled into swappable blocks, letting you replace specific networks without touching the rest. • Smarter default model: The built-in GR00T embodied AI model upgrades from N1.5 to N1.7, offering better action generalization for robots. • Higher PyTorch requirement: Minimum PyTorch version bumped to 2.7 to leverage faster CUDA kernels and improved autograd.