Textual Planning with Explicit Latent Transitions

TL;DR

EmbedPlan introduces a lightweight, embedding-space-based planning method that enables faster multi-step reasoning with LLMs, showing strong within-domain performance but challenges in cross-domain generalization.

Contribution

The paper proposes EmbedPlan, a novel approach that replaces autoregressive generation with a frozen embedding transition model for efficient planning.

Findings

High accuracy in within-domain next-state prediction

Significant performance drop in cross-domain generalization

Effective learning of domain-specific dynamics within a domain

Abstract

Planning with LLMs is bottlenecked by token-by-token generation and repeated full forward passes, making multi-step lookahead and rollout-based search expensive in latency and compute. We propose EmbedPlan, which replaces autoregressive next-state generation with a lightweight transition model operating in a frozen language embedding space. EmbedPlan encodes natural language state and action descriptions into vectors, predicts the next-state embedding, and retrieves the next state by nearest-neighbor similarity, enabling fast planning computation without fine-tuning the encoder. We evaluate next-state prediction across nine classical planning domains using six evaluation protocols of increasing difficulty: interpolation, plan-variant, extrapolation, multi-domain, cross-domain, and leave-one-out. Results show near-perfect interpolation performance but a sharp degradation when generalization requires transfer to unseen problems or unseen domains; plan-variant evaluation indicates generalization to alternative plans rather than memorizing seen trajectories. Overall, frozen embeddings support within-domain dynamics learning after observing a domain's transitions, while transfer across domain boundaries remains a bottleneck.