Search papers, labs, and topics across Lattice.
This paper introduces D-SafeMPC, a novel approach that integrates diffusion models with model predictive control (MPC) to enforce safety and dynamical constraints in robotic planning. By utilizing control barrier functions (CBFs) and control Lyapunov functions (CLFs) within an iterative-projection scheme, D-SafeMPC refines trajectories at each denoising step, ensuring safe and feasible outputs. Experimental results demonstrate significant improvements in safety, task success rates, and planning efficiency on both simulated and real-world robotic tasks compared to existing methods.
D-SafeMPC achieves safer and more efficient robotic planning by seamlessly integrating diffusion models with model predictive control, overcoming key limitations of both approaches.
A key limitation on the use of diffusion models in robotic planning is their inability to inherently enforce safety or dynamical constraints, which often results in physically infeasible or unsafe outputs. Hybrid approaches that employ model predictive control (MPC) to address this problem can be unstable, as poor trajectory initializations from the diffusion model prevent the MPC from converging to a safe and feasible solution. To overcome these challenges, we propose D-SafeMPC, which enhances the interaction between diffusion and control. Our method guides the reverse diffusion process with control barrier functions (CBFs) and control Lyapunov functions (CLFs) and employs an iterative-projection scheme where an MPC refines the trajectory at each denoising step. This steers sampling toward safe, goal-directed regions and provides reliable MPC warm starts. In simulations on a Franka manipulator across four scenarios (one static-obstacle and three dynamic-obstacle settings) and in a sim-to-real experiment on a physical Franka robot, D-SafeMPC improves safety, task success rates, and planning efficiency over state-of-the-art baselines. To facilitate reproducibility, our source code and experimental configurations are available in a repository at https://github.com/erdiphd/D-SafeMPC
