Textual Planning with Explicit Latent Transitions
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914305520173056 |
|---|---|
| author | Shlomi, Eliezer Levy, Ido Shapira, Eilam Katz, Michael Uziel, Guy Shlomov, Segev Mashkif, Nir Reichart, Roi Keren, Sarah |
| author_facet | Shlomi, Eliezer Levy, Ido Shapira, Eilam Katz, Michael Uziel, Guy Shlomov, Segev Mashkif, Nir Reichart, Roi Keren, Sarah |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_04557 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Textual Planning with Explicit Latent Transitions Shlomi, Eliezer Levy, Ido Shapira, Eilam Katz, Michael Uziel, Guy Shlomov, Segev Mashkif, Nir Reichart, Roi Keren, Sarah Computation and Language 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. |
| title | Textual Planning with Explicit Latent Transitions |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.04557 |