Few-shot Subgoal Planning with Language Models
This addresses the challenge of planning in robotics or AI agents with minimal supervision, though it is incremental as it builds on existing language models and benchmarks.
The paper tackled the problem of generating actionable subgoal sequences from text instructions in real-world environments, showing that pre-trained language models can infer detailed subgoal sequences with few training examples and achieve competitive performance on the ALFRED benchmark.
Pre-trained large language models 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 language models allow us to infer fine-grained subgoal sequences. In contrast to recent methods which make strong assumptions about subgoal supervision, our experiments show that language models can infer detailed subgoal sequences from few training sequences without any fine-tuning. We further propose a simple strategy to re-rank language model 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.