OpenRUA: off-the-shelf Claude Code drives robots via ROS 2 at 99% success, zero training
Zhaoyang Chu · hf · 2026-10-06
Prior work on robot-use agents engineers complex custom harnesses and bespoke interfaces. OpenRUA asks whether any of that is necessary—and shows it isn't, with a zero-abstraction design that gives off-the-shelf coding agents nothing but terminal access to the robot's native ROS 2 stack plus docs and basic tools. Perception becomes file I/O; manipulation becomes coding.
- With Claude Code (Claude Opus 5), OpenRUA hits 99.0% on CaP-Bench and 87.0% on LIBERO-PRO—zero-shot visuomotor control with no bespoke primitives or task-specific training
- Emergent behaviors: agents spontaneously write programs to process raw sensor data (96.80% of episodes), build motion-control clients like gripper control (95.87%), and implement closed-loop feedback control (50.13%)
- Code open-sourced at github.com/terminalworld/OpenRUA
More from coding & agent
- Codex spins its wheels while Claude Opus shows far deeper understanding, user reports — iruletheworldmo · 2026-10-06
- Urlbox MCP Server Lets You Capture Screenshots, PDFs and Markdown via Natural Language — modelcontextprotocol · 2026-10-06
- OpenAI Dots Thread Steps 7-8: Approval Gates and Cost-Tiered Intelligence — FinanceYF5 · 2026-10-06
- OpenAI Dots Thread Steps 5-6: Wire in Business Tools and Schedules — FinanceYF5 · 2026-10-06
- 10-Step Blueprint Turns OpenAI Dots Into a 24/7 Company Operations Layer — FinanceYF5 · 2026-10-06
- ODS offers one-click local AI: auto hardware detection, model setup, agents and plugins — Teknium · 2026-10-06