Few-shot Subgoal Planning with Language Models

Abstract: Pre-trained LLMs have shown successful progress in many language understanding benchmarks. This work explores the capability of these models to predict actionable plans in real-world environments. Given a text instruction, we show that language priors encoded in pre-trained LLMs allow us to infer fine-grained subgoal sequences. In contrast to recent methods which make strong assumptions about subgoal supervision, our experiments show that LLMs can infer detailed subgoal sequences from few training sequences without any fine-tuning. We further propose a simple strategy to re-rank LLM predictions based on interaction and feedback from the environment. Combined with pre-trained navigation and visual reasoning components, our approach demonstrates competitive performance on subgoal prediction and task completion in the ALFRED benchmark compared to prior methods that assume more subgoal supervision.