SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts

Published in NeurIPS 2026, 2026

Recommended citation: Raj, A. H., & Xiao, X. (2026). "SCOUT: Planning under Occlusion via Object-Centric World Model Rollouts." NeurIPS 2026. https://arxiv.org/abs/2611.01600

Reconstructs hidden geometry into a relational planning state. An LLM proposes actions and tree search evaluates them with a learned world model, validated on real manipulation in a cluttered shelf.