Search papers, labs, and topics across Lattice.
This paper introduces a Graph Neural Network Model Predictive Control (GNN-MPC) framework for controlling high-dimensional systems like soft robots. The method combines GNN-based dynamics models, which exploit system sparsity, with a condensing algorithm to reduce the computational burden of MPC. Results demonstrate real-time closed-loop control at 100 Hz for systems with up to 1,000 nodes, achieving sub-centimeter accuracy in reference tracking and outperforming baselines by 63.6%.
Control of 1000-node soft robots at 100Hz is now possible thanks to a GNN-MPC framework that leverages GPU parallelization.
The control of high-dimensional systems, such as soft robots, requires models that faithfully capture complex dynamics while remaining computationally tractable. This work presents a framework that integrates Graph Neural Network (GNN)-based dynamics models with structure-exploiting Model Predictive Control to enable real-time control of high-dimensional systems. By representing the system as a graph with localized interactions, the GNN preserves sparsity, while a tailored condensing algorithm eliminates state variables from the control problem, ensuring efficient computation. The complexity of our condensing algorithm scales linearly with the number of system nodes, and leverages Graphics Processing Unit (GPU) parallelization to achieve real-time performance. The proposed approach is validated in simulation and experimentally on a physical soft robotic trunk. Results show that our method scales to systems with up to 1,000 nodes at 100 Hz in closed-loop, and demonstrates real-time reference tracking on hardware with sub-centimeter accuracy, outperforming baselines by 63.6%. Finally, we show the capability of our method to achieve effective full-body obstacle avoidance.
