EmbodiedRSI: Active Continual Robot Learning Through Hypothesis-Guided Co-Evolution

2026-10-07

AI Summary

This work proposes an autonomously evolving agent framework to address the performance degradation of robotic foundation models under environmental shifts, high data collection costs, and the low trial efficiency of existing self-evolution frameworks. Methodologically, it introduces a dual fast-slow system architecture that maintains a hypothesis graph of competing hypotheses to guide active experiment selection. A value-informed mechanism and reward-based hierarchical memory learning are incorporated to optimize physical interaction exploration, while code-skill co-evolution is implemented to facilitate efficient continual learning. The proposed framework significantly outperforms existing baselines on the RoboCasa365 and LIBERO-Pro benchmarks and achieves high-success-rate zero-shot transfer in real-world robotic tasks.

Abstract

Robot foundation models provide strong visuomotor control, yet their performance can degrade when object positions or task instructions change. Further improvements often require post-training on substantial robot data, which can be costly to collect through methods such as teleoperation. Agentic harnesses can adapt around the model, but current self-evolving harnesses use robot trials inefficiently when deciding which code and skill changes to pursue. We introduce EmbodiedRSI, a self-evolving agentic harness that autonomously decides where to explore next and turns the resulting physical interaction into improved code and skills. EmbodiedRSI realizes this through a Fast-Slow Dual-System Architecture, in which competing code and skill hypotheses are maintained in a Hypothesis Graph. Value-of-Information Experiment Selection chooses physical experiments that can distinguish these hypotheses. Their outcomes guide Code-Skill Co-Evolution. The Slow System builds Hierarchical Memory, and Reward-Grounded Memory Learning selects effective memory according to their value for later Fast-System improvement. On RoboCasa365, EmbodiedRSI reaches 77.0% overall success and 71.3% on Composite-Unseen, compared with 40.1% for the best baseline. EmbodiedRSI also reaches 86.8% overall success on LIBERO-Pro. Beyond benchmark performance, EmbodiedRSI transfers zero-shot to real-world robot, achieving 71.3% overall success across multiple challenging tasks.

Problem

• Robot Foundation Models
• Continual Learning
• Self-Evolving Agentic Harness
• Sample Efficiency

Innovation

• Self-evolving agentic harness
• Fast-Slow Dual-System Architecture
• Hypothesis Graph
• Code-Skill Co-Evolution
• Continual Robot Learning