Due to its factorized architecture, Diffusion Model Predictive Control (DMPC) can adapt to changes in an environment's dynamics by re-training only its dynamics model component on new data.

A key advantage of Model Predictive Control (MPC) is its ability to adapt to new reward functions at test time without retraining.

The Diffusion Policy model for robotics requires expert demonstrations to train and operates within a behavior cloning paradigm.

The Decision Diffuser architecture allows for learning from video-only data, which is a key benefit for robotics where collecting action-labeled data is a bottleneck.

Diffusion Model Predictive Control (DMPC) can outperform previous approaches using a simple sampling-based planner due to the powerful modeling capabilities of diffusion models.

Diffusion Model Predictive Control (DMPC) uses diffusion models to learn both multi-step action proposals and multi-step dynamics models for robotic control.