Decision tree learns from a language model to help robots avoid freezing
Teaching a Tree to Recover: Online Distillation of an LLM Supervisor with Counterfactual Replay
Robotics
Summary
Autonomous robots sometimes get stuck because their quick obstacle-avoiding plans don't always work perfectly. The authors developed a way to teach a simple decision tree using advice from a large language model (LLM) so the robot can decide when to replan, wait, or ask for help without slowing down. The decision tree improves over time by learning from past experiences and only asks the LLM when facing new problems. This approach helps robots handle tricky situations more reliably without always needing the slower language model.
What this means in practice
- •For robotics engineers: Enable robots to handle complex navigation failures by relying on a learned decision tree instead of repeatedly querying costly language models.
- •For robotics operations teams: Implement supervisory control that safely calls for help only when no feasible path exists, minimizing unnecessary interventions during robot navigation.
Authors
Kristian Ceder, Kilian Freitag, Knut Åkesson
Abstract
Autonomous mobile robots commonly pair a global planner with a local model predictive controller (MPC), but nonconvex obstacle avoidance and a limited horizon can make the robot freeze in complex layouts. Deciding whether to replan, wait, or call for help is a supervisory decision. Recent work queries a large language model (LLM) supervisor on every anomaly, which is slow and costly. We instead distill the LLM online into a decision tree. After each training episode, counterfactual replay labels recoveries as needed, unnecessary, or harmful. These labels retire unhelpful rules. In simulation, on 20 unseen layouts where the MPC alone freezes, the tree handles at least 18 correctly across all seeds, either by reaching the goal or by calling for help when no feasible path exists. The tree matches both an always-query LLM supervisor and expert-designed rules, while issuing no LLM queries at test time. Handwritten rules cannot anticipate the long tail of situations a physical robot will encounter. Online distillation is a promising remedy: the tree handles familiar cases, and the LLM is consulted only for new ones, whose validated recoveries become new rules. Language models can thus also serve as designers of supervisors rather than only as runtime supervisors.