SERA self-evolving harness lifts long-horizon robot planning on RoboMME
Original title: From VLM Planners to Robotic Agents: Self-Evolving Harnesses for History-Dependent Manipulation
With a fixed VLA executor, autonomous harness selection reached 51.38% success versus 44.38% without self-evolution.
SERA organizes a high-level vision-language model as a robotic agent inside an editable harness, while a fixed vision-language-action model handles low-level control. The harness keeps task context from observations and execution feedback so planning can depend on past progress. Environment rollouts propose and score harness revisions, and the chosen setup is frozen before held-out tests. On RoboMME, autonomous selection reached 51.38% overall success.
- Success was 51.38% with self-evolution and 44.38% without it.
- The selected harness configuration is fixed before held-out testing.
- The harness updates task context from observations and execution feedback.
From the original article
1 min read at huggingface.co
Vision-language-action (VLA) models achieve strong performance across embodied manipulation tasks by unifying visual understanding, language conditioning, and action generation. However, long-horizon tasks can involve non-Markovian dependencies: successful decisions may require retaining task progress and past interaction outcomes, while partial observability and execution errors compound over time.
By Tingwen Du, Haoyu Guo, Hanxue Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.