Policy Adaptation via Language Optimization: Decomposing Tasks for Few-Shot Imitation

Vivek Myers, Chunyuan Zheng, Oier Mees, Kuan Fang, Sergey Levine

Keywords: Reinforcement Learning, Vision-Language Models, Manipulation

TL;DR: We show decomposing tasks with language can enable few-shot adaptation to OOD manipulation tasks.

Abstract: Learned language-conditioned robot policies often struggle to effectively adapt to new real-world tasks even when pre-trained across a diverse set of instructions. We propose a novel approach for few-shot adaptation to unseen tasks that exploits the semantic understanding of task decomposition provided by vision-language models (VLMs). Our method, Policy Adaptation via Language Optimization (PALO), combines a handful of demonstrations of a task with proposed language decompositions sampled from a VLM to quickly enable rapid nonparametric adaptation, avoiding the need for a larger fine-tuning dataset. We evaluate PALO on extensive real-world experiments consisting of challenging unseen, long-horizon robot manipulation tasks. We find that PALO is able of consistently complete long-horizon, multi-tier tasks in the real world, outperforming state of the art pre-trained generalist policies, and methods that have access to the same demonstrations.

Video: https://www.youtube.com/watch?v=mRJcocqKe6g&feature=youtu.be

Website: https://palo-website.github.io

Code: https://github.com/vivekmyers/palo-robot

Student Paper: yes

Submission Number: 698