Event
Proposes physical RSI evaluation PhysRSI-eval and studies observational scaling laws
1 reports · 1 sources
The authors of the PhysRSI-eval paper propose quantifying the physical recursive self-improvement (RSI) of embodied agents using five dimensions: first-attempt performance, efficiency, success rate, budget, and failure improvement, and evaluate the performance of multiple VLMs on 18 manipulation tasks under a fixed interaction budget. The paper analyzes observational scaling and workspace intervention experiments: removing history reduces the success rate in later rounds, a frozen workspace remains useful on unseen layouts, and three cross-task cases achieved success earlier.
Report timeline
- PhysRSI-eval: An Observational Scaling Law Study of Physical Recursive Self-Improvement
This paper systematically studies the mechanisms, scaling behavior, and generalization of physical recursive self-improvement (RSI) in embodied agents, proposing PhysRSI-eval, which quantifies physical RSI across five dimensions: first-attempt performance, efficiency, success rate, budget, and improvement from failure, and evaluates multiple VLMs on 18 manipulation tasks under a fixed interaction budget.
HuggingFace Daily Papers ·