Newest papers

Where is Physical RSI capability from? An Observational Scaling Law of LLM-as-policy

Zekai Li, Jiayi Yu, Jinlu Zhang, Luyang Niu, Zekun Qi, Jun Zheng, Xuchuan Chen, Dairu Liu, Yintianrun Zhang, He Wang, Li Yi · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26848 · cited by 0 works (OpenAlex) · open-access PDF

French

Dans ce document, nous enquêtons systématiquement sur les mécanismes, le comportement d'échelle et la généralisation de l'auto-amélioration récurrente physique. Nous introduisons PhysRSI-eval pour quantifier le RSI physique à travers une métrologie à cinq axes: Première tentative de performance, efficacité, succès, budget et amélioration par rapport à l'échec. Les différents VLM sont évalués sur 18 tâches de manipulation sous un budget d'interaction fixe. L'analyse des résultats expérimentaux motive l'hypothèse de Barrel physique: différentes capacités peuvent restreindre différentes parties de l'amélioration. Les métriques à cinq axes distinguent également les systèmes avec des moyennes de référence publiques étroites et montrent que des scores élevés

Machine translation (Meta M2M100); the English original follows.

Abstract

In this paper, we systematically investigate the mechanisms, scaling behavior, and generalization of physical Recursive self-improvement. We introduce PhysRSI-eval to quantify physical RSI through a five-axis metric: First-Attempt Performance, Efficiency, Success, Budget and Improvement from failure. Different VLMs are evaluated on 18 manipulation tasks under a fixed interaction budget. The analysis of experimental results motivates the Physical Barrel Hypothesis: different capabilities may constrain different parts of improvement. The five-axis metrics also distinguish systems with close public benchmark averages and show that high scores need not come with early or efficient gains. We study observational scaling from two perspectives: predicting the relative performance of VLM agents from their public capabilities, and forecasting later progress from early physical feedback. Together, these analyses ask how quickly and how far an agent can improve. Moreover, we conduct workspace interventions and then analyze what improvement leaves behind. Removing history reduces late-round success, frozen workspaces remain useful on unseen layouts, and three cross-task cases achieve success earlier. The results suggest that scene-specific adaptation and reusable experience can coexist.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Benchmark (surveying), Observational study, Workspace, Computer science, Scaling, Generalization

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange