Learning to Learn Faster from Human Feedback with Language Model Predictive Control

TL;DR

This paper introduces Language Model Predictive Control (LMPC), a fine-tuning approach for LLMs that enhances robot code learning from human feedback by modeling interactions as a decision process, leading to improved teachability and success rates.

Contribution

The paper proposes LMPC, combining fine-tuned LLMs with model predictive control to better remember interactions and improve robot teaching efficiency across multiple tasks and embodiments.

Findings

26.9% increase in success rate for unseen tasks

Reduced human corrections from 2.4 to 1.9 on average

31.5% improvement in in-context learning success rate

Abstract

Large language models (LLMs) have been shown to exhibit a wide range of capabilities, such as writing robot code from language commands -- enabling non-experts to direct robot behaviors, modify them based on feedback, or compose them to perform new tasks. However, these capabilities (driven by in-context learning) are limited to short-term interactions, where users' feedback remains relevant for only as long as it fits within the context size of the LLM, and can be forgotten over longer interactions. In this work, we investigate fine-tuning the robot code-writing LLMs, to remember their in-context interactions and improve their teachability i.e., how efficiently they adapt to human inputs (measured by average number of corrections before the user considers the task successful). Our key observation is that when human-robot interactions are viewed as a partially observable Markov decision process (in which human language inputs are observations, and robot code outputs are actions), then training an LLM to complete previous interactions is training a transition dynamics model -- that can be combined with classic robotics techniques such as model predictive control (MPC) to discover shorter paths to success. This gives rise to Language Model Predictive Control (LMPC), a framework that fine-tunes PaLM 2 to improve its teachability on 78 tasks across 5 robot embodiments -- improving non-expert teaching success rates of unseen tasks by 26.9% while reducing the average number of human corrections from 2.4 to 1.9. Experiments show that LMPC also produces strong meta-learners, improving the success rate of in-context learning new tasks on unseen robot embodiments and APIs by 31.5%. See videos, code, and demos at: https://robot-teaching.github.io/.