Text2Motion: From Natural Language Instructions to Feasible Plans

Kevin Lin, Christopher Agia, Toki Migimatsu, Marco Pavone, Jeannette Bohg

Keywords: Long horizon planning, Task and Motion Planning, Large Language Models, Grounding Models

TL;DR: Long horizon robot plans generated by large language models can be optimized and verified for instruction satisfaction using dynamics models, Q functions and optimization.

Abstract:

We propose Text2Motion, a language-based planning framework enabling robots to solve sequential manipulation tasks that require long-horizon reasoning. Given a natural language instruction, our framework constructs both a task- and motion-level plan that is verified to reach inferred symbolic goals. Text2Motion uses feasibility heuristics encoded in Q-functions of a library of skills to guide task planning with Large Language Models. Whereas previous language-based planners only consider the feasibility of individual skills, Text2Motion actively resolves geometric dependencies spanning skill sequences by performing geometric feasibility planning during its search.
We evaluate our method on a suite of problems that require long-horizon reasoning, interpretation of abstract goals, and handling of partial affordance perception. Our experiments show that Text2Motion can solve these challenging problems with a success rate of 82\%, while prior state-of-the-art language-based planning methods only achieve 13\%. Text2Motion thus provides promising generalization characteristics to semantically diverse sequential manipulation tasks with geometric dependencies between skills.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2303.12153/code)