Sim-to-Real Benchmarks, Open-Source Arms, and the Push for Reliable Manipulation

Research

REACT: Real-World Evaluation for Autonomous Contact-rich Tasks

This benchmark directly measures contact-rich manipulation in the real world, not just in simulation. For builders, it's a much-needed reality check on how well your force-control policies actually transfer.

Bench2Free: Benchmarking Object Manipulation from Free-Form to Goal-Oriented Tasks

Another new benchmark, but this one focuses on the messy, free-form tasks you find outside the lab. It's pushing the field beyond rigid pick-and-place to assess true generalization.

Distilling Vision-Language-Action Models for Efficient Robotic Control

This work tackles the main bottleneck in VLA deployment: inference speed. Distillation to smaller, faster models is the pragmatic path to getting these large models onto real robots in production.

Tools

Afford-X: An Open-Source Anthropomorphic Robotic Arm with Multi-Modal Sensing

This is a full open-source hardware release for a dexterous arm with integrated tactile sensing. It lowers the barrier for teams to prototype and test contact-rich manipulation algorithms without custom hardware.

NVIDIA Isaac Sim Updates for High-Fidelity Data Generation

NVIDIA keeps pushing Isaac Sim as the default factory for synthetic training data. The latest updates focus on closing the sim-to-real gap for manipulation, which is critical for scaling policy training.

Events

ROSCon 2025 Call for Proposals and Program Updates

ROSCon is the core developer conference for the ROS ecosystem. Watching the accepted talks and workshops gives you a clear map of where the open-source robotics stack is heading next.