Planning with Generative Cognitive Maps

Jeffrey Qin, Albert Yang, Cole Wyeth, Ziheng Xu, Kevin Ellis, Marta Kryven

Keywords: planning, cognitive maps, cognitive science, large language models, world-modelling with llm, human-cenetered AI

TL;DR: We show that people approximate planning by piecewise policies conditioned on world structure, and contribute a proof-of concept framework that implement these principles in a working system by using world-modelling with llm.

Abstract:

Planning relies on cognitive maps -- models that encode world structure given cognitive resource constraints. The problem of learning functional cognitive maps is shared by humans, animals and machines.
However, we still lack a clear understanding of how people represent maps for planning, particularly when the goal is to support cost-efficient plans.
We take inspiration from theory of compositional mental representations in cognitive science to propose GenPlan: a cognitively-grounded computational framework that models redundant structure in maps and saves planning cost through policy reuse. Our framework integrates (1) a Generative Map Module that infers generative compositional structure and (2) a Structure-Based Planner that exploits structural redundancies to reduce planning costs. We show that our framework closely aligns with human behaviour, suggesting that people approximate planning by piece-wise policies conditioned on world structure. We also show that our approach reduces the computational cost of planning while producing good-enough plans, and contribute a proof-of-concept implementation demonstrating how to build these principles into a working system.

Submission Type: Research Paper (4-9 Pages)

Submission Number: 88