Frozen skills cut costs and boost reliability in robot tasks
Compile Once, Run Cheap: Frozen, Portable Skills for Embodied Agents, and What Recursive Self-Improvement Adds
RoboticsArtificial IntelligenceMachine Learning
Summary
Teaching robots new skills usually takes many tries and lots of data, making it expensive and slow. This paper shows a way to create robot skills once using a strong model that writes and fixes the skill in simulation until it works reliably. The saved skill can then be run cheaply many times without needing the powerful model again. The authors found that skills improved by reusing past versions worked better, and frozen skills performed well on new tasks with very low running costs.
What this means in practice
- •For robotics engineers: Create reusable, tested robot manipulation skills offline to run cheaply and reliably in multiple environments without retraining each time.
- •For automation system developers: Deploy embodied AI tasks where frozen skills cut runtime costs and reduce the need for continuous model access, improving operational efficiency.
Authors
Jingwen Bai, Guochao Bai, Canbo Zhang, Shuangqi Li, Minchao Hu, Weichun Huang, Shuang Liu
Abstract
Embodied AI lacks the data-rich pretraining that gave language models their agents and skills, so new manipulation capability costs demonstrations per task or a frontier model in every episode. We study a third route: a strong general model acts only as an offline author that writes, runs, diagnoses and repairs a manipulation skill in simulation until it passes an exam on layouts drawn after freezing; the frozen skill then runs as code alone or under a cheap caller that reads only its description. We call this skill-level recursive self-improvement Skill-RSI and specify a compile protocol (isolated, audited author; hidden validation layouts; no-feedback and inherit-only controls; tamper-evident ledger) and a reuse protocol that runs the identical files with executors from another model family. On 7 RoboTwin 2.0 tasks with cluttered tables, Claude Code compiled 4 tasks into 7 frozen skills, 5 from a first draft needing only qualification, at USD 4.29-8.22 of author cost each (USD 55 per skill counting the 13 failed campaigns). The other three tasks did not compile in 18 campaigns, including five with GPT-6 and a pre-registered 150-layout run: on the hardest task every campaign scored 78-95 of 100 but had 3-16 safety aborts, failing the gate of 95% success with zero aborts; a 100-episode exam bounds an abort rate only to 3.6%. On hidden layouts, repair that inherits the previous version beat from-scratch rewriting in 8 of 8 controlled campaigns (sign test p = 0.004); an inherit-only lineage attributes most of the gap to inheritance, with development feedback adding a little (3 of 3, p = 0.125). Write-once campaigns found no detectable first-draft effect of practice (minimum detectable effect 14 points). The frozen skills kept 148/150 successes on new layouts with no model and 147/150 with a DeepSeek-flash caller at USD 0.00028-0.00054 per episode; no direct-control baseline was run.