Differentiable MPC for End-to-end Planning and Control
This work addresses data efficiency and expert unrealizability in continuous control for robotics and autonomous systems, though it is incremental as it builds on existing MPC and differentiable optimization methods.
The paper tackled the problem of combining model-free and model-based reinforcement learning by introducing differentiable MPC for end-to-end planning and control, showing that MPC policies are significantly more data-efficient than generic neural networks and superior to traditional system identification in unrealizable expert settings.
We present foundations for using Model Predictive Control (MPC) as a differentiable policy class for reinforcement learning in continuous state and action spaces. This provides one way of leveraging and combining the advantages of model-free and model-based approaches. Specifically, we differentiate through MPC by using the KKT conditions of the convex approximation at a fixed point of the controller. Using this strategy, we are able to learn the cost and dynamics of a controller via end-to-end learning. Our experiments focus on imitation learning in the pendulum and cartpole domains, where we learn the cost and dynamics terms of an MPC policy class. We show that our MPC policies are significantly more data-efficient than a generic neural network and that our method is superior to traditional system identification in a setting where the expert is unrealizable.