Dream-MPC: Gradient-Based Model Predictive Control with Latent Imagination

arXiv 2026 55.4 method

TLDR

Dream-MPC combines gradient-based MPC with a learned world model and policy prior, outperforming gradient-free methods on 24 continuous control tasks.

Reasoning

The paper presents a novel gradient-based MPC approach that leverages a learned world model for trajectory optimization, showing strong empirical results across many tasks. However, it lacks real-world validation and does not address video or interactive world models explicitly.

Read-first score

Read-first score 55.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

72.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.72374803

Topical relevance 29%

61.4

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 18%

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

Reproducibility 18%

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

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierBridge

Rank sensitivity

Stability: volatile; rank range: 289.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

generative world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Gradient-based Model Predictive Control with latent imagination using a learned world model
Uncertainty regularization for trajectory optimization
Amortization of optimization iterations over time by reusing previously optimized actions

Methodology

Dream-MPC generates a small number of candidate trajectories by rolling out a policy, then optimizes each trajectory via gradient ascent using a learned world model. It incorporates uncertainty regularization and amortizes optimization iterations across time steps by reusing previously optimized actions.

Key Results

On 24 continuous control tasks, Dream-MPC significantly improves the performance of the underlying policy and outperforms gradient-free MPC as well as state-of-the-art baselines.

Resources

Paper