ARC: A Reasoning Recipe for Robot Foundation Models

Key points

• The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.1
• First, we find that effective reasoning traces should be grounded in the robot's next action and explain its causal structure: why the action is appropriate and what effect it should produce.1
• Second, we show that these traces can be generated automatically from existing demonstrations, enabling us to construct ARC-Trace-DROID from DROID without collecting new robot data.1
• Third, we show how state-of-the-art VLAs such as $π{0.5}$ and WAMs such as Cosmos3-Nano-Policy can learn to use these traces for control, with fine-tuning and inference tailored to each model's architecture and capabilities.1

Sources (1)

• [1]ARC: A Reasoning Recipe for Robot Foundation Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 05:38 PM

  We show that there exists an effective and efficient complementary approach: the right reasoning recipe can substantially improve the zero-shot task performance of existing state-of-the-art RFMs. We refer to this recipe as ARC.

  The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.

Extractive summary: sentences quoted from the sources.