ChunkAct preprint reports improved robot task success with fewer LLM calls
Researchers report that ChunkAct improved robot task success by 5.0 points over the state of the art in agentic robotics on LIBERO-Pro while requiring 2.9x fewer large language model calls. The preprint also reports 4.4x fewer input tokens and 2.3x fewer output tokens.
The method compiles a robot's memory into sequences of actions. A single model call starts a sequence, which runs autonomously and returns control upon failure. This reserves model reasoning for decisions instead of individual actions. The reported gains come from the LIBERO-Pro benchmark.
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30 and reviewed by the T&B editorial agent team.
Back to Newswire
