Saved in:
Bibliographic Details
Main Authors: Gao, Baoqi, Han, Ruize, Wang, Miao, Wang, Song
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2606.01626
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918534526795776
author Gao, Baoqi
Han, Ruize
Wang, Miao
Wang, Song
author_facet Gao, Baoqi
Han, Ruize
Wang, Miao
Wang, Song
contents Planning with a learned latent world model is a promising route to control from raw pixels, but a strong world model alone is not enough. We show this experimentally: even with a perfect world model (operationalized by replacing the learned forward predictor with an idealized rollout of the true environment dynamics), a finite-budget sample-based planner still fails on some tasks, indicating that the bottleneck can lie in search rather than in world-model accuracy. Motivated by this gap, we propose IMWM (Intuition Model + World Model), which pairs the world model with an intuition model trained from demonstrations to recognize promising actions. The two models collaborate through three lightweight components: (i) Retrieval Initialization, which initializes the planner's action proposal from a retrieved demonstration; (ii) Hybrid Cost, which combines the intuition score with the world-model rollout cost; and (iii) a Reliability Gate, which adjusts how much the planner trusts intuition in each setting. Across four pixel-based goal-reaching tasks (Two-Room, Reacher, Push-T, and OGBench-Cube), IMWM has higher mean success than the world-model-only planner on all four, with the largest gains on Two-Room (99.2%, +11.5 percentage points) and OGBench-Cube (94.7%, +28.5 percentage points).
format Preprint
id arxiv_https___arxiv_org_abs_2606_01626
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle IMWM: Intuition Models Complement World Models for Latent Planning
Gao, Baoqi
Han, Ruize
Wang, Miao
Wang, Song
Machine Learning
Planning with a learned latent world model is a promising route to control from raw pixels, but a strong world model alone is not enough. We show this experimentally: even with a perfect world model (operationalized by replacing the learned forward predictor with an idealized rollout of the true environment dynamics), a finite-budget sample-based planner still fails on some tasks, indicating that the bottleneck can lie in search rather than in world-model accuracy. Motivated by this gap, we propose IMWM (Intuition Model + World Model), which pairs the world model with an intuition model trained from demonstrations to recognize promising actions. The two models collaborate through three lightweight components: (i) Retrieval Initialization, which initializes the planner's action proposal from a retrieved demonstration; (ii) Hybrid Cost, which combines the intuition score with the world-model rollout cost; and (iii) a Reliability Gate, which adjusts how much the planner trusts intuition in each setting. Across four pixel-based goal-reaching tasks (Two-Room, Reacher, Push-T, and OGBench-Cube), IMWM has higher mean success than the world-model-only planner on all four, with the largest gains on Two-Room (99.2%, +11.5 percentage points) and OGBench-Cube (94.7%, +28.5 percentage points).
title IMWM: Intuition Models Complement World Models for Latent Planning
topic Machine Learning
url https://arxiv.org/abs/2606.01626