Who Improves the Robot? Toward Physical Recursive Self-Improvement

A research paper from Shanghai Jiao Tong University, UC Berkeley, Stanford, MIT, Google, Microsoft, NVIDIA, and Xiaomi introduces Physical Recursive Self-Improvement (Physical RSI), a framework for enabling robots to autonomously update their task execution systems and the processes that drive their improvements. The paper defines L0–L5 levels to assess the degree of self-improvement, outlines challenges in physical environments, and proposes a unified framework for evaluating recursive self-improvement in robotics.

A research paper proposes Physical RSI, a framework enabling robots to autonomously update their task execution systems and improvement processes using physical interaction experience. The paper introduces a L0–L5 grading system to evaluate self-improvement levels and highlights challenges in achieving recursive gains in real-world environments, including safety, validation, and cross-task generalization.

A research paper explores recursive self-improvement in physical robots, proposing a unified framework for understanding how robots can autonomously update their capabilities and improvement processes. The study defines L0–L5 levels to evaluate the effectiveness of self-improvement methods, identifying gaps in current research that prevent robots from continuously learning across tasks.