Harvard SEAS Develops Diffusion-MPC for Adaptive Legged Robots
A team at Harvard's John A. Paulson School of Engineering and Applied Sciences (SEAS) led by Yilun Du has developed Diffusion-MPC (Model Predictive Control), an AI-based control system for four-legged robots. This system allows robots to quickly adapt movements to new tasks and terrains without retraining, using a generative diffusion model to predict robot and environment evolution. It refines tasks in real time by rewarding good performance and enforcing constraints.
Diffusion-MPC represents a significant step towards general-purpose locomotion controllers, allowing robots to adjust to changing conditions in the field rather than being limited to specific trained scenarios, enhancing their real-world applicability.
Learn one new AI thing every day.
Daily Deck sends you seven plain-English cards like this every morning. Free.
Start free