PhysEvo: Astra Can Act, Let It
PhysEvo: Astra Can Act, Let It
This paper presents PhysEvo, a framework for physical recursive self-improvement (RSI) that allows a frozen model to adapt and improve through task execution and meta-agent feedback. The framework enables joint-level control, evidence-seeking observation, and reusable manipulation skills without model-weight updates or a separately trained action policy. Evaluated on 42 RoboDojo tasks and 8 manipulation tasks, PhysEvo achieves an average score of 68.14/100 and 62.00% success, significantly outperforming existing methods. Deployment on real-world tasks shows 84.00% success across 25 trials.
PhysEvo introduces a framework for physical recursive self-improvement (RSI) that enables a frozen model to adapt and improve through task execution and meta-agent feedback, achieving higher success rates compared to existing methods like RoboDawn's Astra agent.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
average score
90.6
View original evidence
-dimension average score of 68.14/100 and 62.00% success, compared with 47.17% for RoboDawn's one-shot Astra agent, the strongest published reference in our comparison. On eight manipulation tasks challenging direct Astra, PhysEvo achieves 55.00% success, compared with 1.25% for the direct-Astra reference. Deploying the simulation-evolved harness on AgileX PiPER and continuing skill revision yields 90.60/100 average
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
The meta-agent can also improve its own diagnostic tools, so retained revisions support both later action and later self-improvement. This process develops joint-level control, evidence-seeking observation, and reusable manipulation skills without model-weight updates or a separately trained action policy. Across 42 RoboDojo tasks, held-out-layout evaluation of retained task-specific deployment versions yields a five
Open source S3
-dimension average score of 68.14/100 and 62.00% success, compared with 47.17% for RoboDawn's one-shot Astra agent, the strongest published reference in our comparison. On eight manipulation tasks challenging direct Astra, PhysEvo achieves 55.00% success, compared with 1.25% for the direct-Astra reference. Deploying the simulation-evolved harness on AgileX PiPER and continuing skill revision yields 90.60/100 average
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org