AI

Learning How to Search for Plans with Exponentially Less Space

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 04:03:25 UTC.

Briefings

This article presents a breakthrough in planning search where the system learns a policy that uses polynomial space instead of exponential space, even when the heuristic is nearly perfect. The method uses a counterexample-guided loop with a language model to learn an indexical policy with registers and modes. This approach solves 1,709 out of 1,890 test tasks from the IPC 2023 Learning Track and the Autoscale Agile suite, outperforming previous methods like LAMA, BFWS, and Levitron, with most tasks completed in under one second and using less than 100 MiB of memory.

Key points

• The system learns an indexical policy with registers and modes that bounds every execution by a polynomial in the number of objects, achieving structural termination.
• The method uses a counterexample-guided loop with a language model to learn policies that solve 1,709 of 1,890 test tasks from IPC 2023 and Autoscale Agile within one second and 100 MiB of memory.

Context and limitations

The study covers 1,709 of 1,890 test tasks from IPC 2023 and Autoscale Agile; results are based on a counterexample-guided loop with a language model and may vary with different datasets or configurations.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/10/2026, 04:42:17 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeigz3ccitgbwiylf67abbvdc5jqfyuk6juvpiu6npdgwvdgrua6eae