PyRUA-REAL: coding agents give robots recursive self-improvement in the real world

Qiang Wang · hf · 2026-11-05

A paper from Peking University, EngineAI, and NUS introduces PyRUA-REAL, a framework for embodied recursive self-improvement (RSI) in the real world.

Coding agents powered by models like GPT-6-Astra connect spatial reasoning with executable robot behavior
Physical interaction experience is retained as persistent memory and reusable executable skills
The loop: real-world exploration → code refinement → deployment, with the skill library guiding further improvement
Achieves high success rates on both real-world and benchmark tasks

Notable for grounding the "agents crystallize experience into code skills" idea on physical robots.

More from Embodied