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

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 formulated 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 can be viewed as 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/.

Authors

Ken Caluwaerts, Ben Jyenis, Jasmine Hsu, andyzeng , Wenhao Yu, Nik Stewart, Jacky Liang, Fei Xia, Peng Xu, Jie Tan, ichter , Erik Frey, Carolina Parada, Dorsa Sadigh, Tingnan Zhang, Ted Xiao, Zhuo Xu, Nikhil Joshi, Kuang-Huei Lee, Chase Kew, Ken Oslund, Sean Kirmani, Dushyant Rao, quanhovuong , Keerthana Gopalakrishnan, Marissa Giustina, Jonathan Tompson, Assaf Hurwitz Michaely, Baruch Tabanpour, Maria Bauza, Edward Lee, Maria Attarian, Leonard Hasenclever, Alex Bewley, Jan Humplik, Nimrod Gileadi, Joss Moore, Leila Takayama, allenren , Adil Dostmohamed, Chuyuan Kelly Fu, Ayzaan Wahid, Matt Bennice, Vincent Zhuang, Nicolas Heess, Izhak Shafran, Vincent Vanhoucke, Maja Mataric, Montse Gonzalez Arenas, Ying Xu, Kanishka Rao

Venue

arXiv