Stein Variational Model Predictive Control
This addresses the problem of handling complex probability distributions in MPC for autonomous systems, representing an incremental improvement by integrating variational methods into an existing framework.
The paper tackles decision-making under uncertainty in autonomous systems by proposing a generalization of Model Predictive Control (MPC) that represents solutions as posterior distributions, casting MPC as a Bayesian inference problem and using Stein variational gradient descent to estimate these distributions, resulting in successful planning in challenging, non-convex optimal control problems.
Decision making under uncertainty is critical to real-world, autonomous systems. Model Predictive Control (MPC) methods have demonstrated favorable performance in practice, but remain limited when dealing with complex probability distributions. In this paper, we propose a generalization of MPC that represents a multitude of solutions as posterior distributions. By casting MPC as a Bayesian inference problem, we employ variational methods for posterior computation, naturally encoding the complexity and multi-modality of the decision making problem. We present a Stein variational gradient descent method to estimate the posterior directly over control parameters, given a cost function and observed state trajectories. We show that this framework leads to successful planning in challenging, non-convex optimal control problems.