PhysEvo: Astra Can Act, Let It

Key points

• Astra can act, yet reliable manipulation depends on the system through which it observes and controls the world.1
• A task agent executes robot tasks; a meta-agent uses the resulting trajectories to diagnose failures, revise tools and skills, and test corrections.1
• Across 42 RoboDojo tasks, held-out-layout evaluation of retained task-specific deployment versions yields a five-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.1
• On eight manipulation tasks challenging direct Astra, PhysEvo achieves 55.00% success, compared with 1.25% for the direct-Astra reference.1

Sources (1)

• [1]PhysEvo: Astra Can Act, Let It
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 06:54 PM

  We introduce PhysEvo, a framework for physical recursive self-improvement (RSI) around a single frozen model.

  Astra can act, yet reliable manipulation depends on the system through which it observes and controls the world.

Extractive summary: sentences quoted from the sources.