ChronoDreamer: Action-Conditioned World Model as an Online Simulator for Robotic Planning

arXiv 25.12 2025 68.8 method

TLDR

ChronoDreamer is an action-conditioned world model for robotic manipulation that predicts future video and contact using a spatial-temporal transformer and a VLM for safe planning.

Reasoning

The paper introduces a novel contact representation via depth-weighted Gaussian splat images and integrates a vision-language model for collision-aware rejection sampling, which are strengths. However, it lacks real-world experiments and quantitative evaluation, relying solely on simulation and qualitative results.

Read-first score

Read-first score 68.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 59.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

84.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,result

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 215.

Keyword Scores

world model

world simulator

interactive world model

video world model

generative world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Action-conditioned world model for contact-rich robotic manipulation using egocentric RGB, contact maps, actions, and joint states
Depth-weighted Gaussian splat images to encode 3D contact forces into a camera-aligned format compatible with vision backbones
Spatial-temporal transformer trained with MaskGIT-style masked prediction for joint prediction of video, contact distributions, and joint angles
LLM-based judge for rejection sampling of unsafe actions by evaluating collision likelihood from predicted rollouts

Methodology

ChronoDreamer uses a spatial-temporal transformer trained with MaskGIT-style masked prediction to jointly predict future video frames, contact distributions, and joint angles from a history of egocentric RGB frames, contact maps, actions, and joint states. Contact is encoded as depth-weighted Gaussian splat images that render 3D forces into a camera-aligned format. At inference, a vision-language model evaluates predicted rollouts for collision likelihood to perform rejection sampling of unsafe actions. The model is trained and evaluated on DreamerBench, a simulation dataset generated with Project Chrono providing synchronized RGB, contact splat, proprioception, and physics annotations.

Key Results

Qualitative results show that the model preserves spatial coherence during non-contact motion and generates plausible contact predictions, while the LLM-based judge effectively distinguishes collision from non-collision trajectories.

Resources

Paper