Learning Bilevel Policies over Symbolic World Models for Long-Horizon Planning
The paper discusses a new approach to long-horizon planning in embodied AI agents. It combines low-level imitation learning with high-level symbolic abstractions to create bilevel policies. The proposed BISON system demonstrates improved efficiency and generalization in solving complex tasks.
- ▪The research addresses the challenge of long-horizon planning for AI agents.
- ▪Bilevel policies are introduced, consisting of a neural policy for low-level tasks and a symbolic policy for high-level planning.
- ▪Experiments show that BISON can efficiently solve problems with a large number of objects.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.15975 (cs) [Submitted on 15 May 2026] Title:Learning Bilevel Policies over Symbolic World Models for Long-Horizon Planning Authors:Dillon Z. Chen, Till Hofmann, Toryn Q. Klassen, Sheila A. McIlraith View a PDF of the paper titled Learning Bilevel Policies over Symbolic World Models for Long-Horizon Planning, by Dillon Z. Chen and Till Hofmann and Toryn Q. Klassen and Sheila A. McIlraith View PDF HTML (experimental) Abstract:We tackle the challenge of building embodied AI agents that can reliably solve long-horizon planning problems.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
