Automatic Tuning for Data-driven Model Predictive Control

Model predictive control (MPC) is a powerful feedback technique that is often used in data-driven robotics. The performance of data-driven MPC depends on the accuracy of the model, which often requires careful tuning. Furthermore, specifying the task with an objective function and synthesizing a feedback policy are not straightforward and typically lead to suboptimal solutions driven by trial and error. To address these challenges, we present a method to jointly optimize the data-driven system identification, task specification, and control synthesis of unknown dynamical systems. We use our method to develop AutoMPC 3, a software package designed to automate and optimize data-driven MPC. Empirical evaluation on the pendulum swing-up, cart-pole swing-up, and half-cheetah running demonstrates that our method finds data-driven control policies that outperform offline reinforcement learning, without any hand-tuning.

Authors

• William Edwards
• Gao Tang
• Giorgos Mamakoukas
• Todd D. Murphey
• Kris Hauser

Keywords

• Software packages
• Conferences
• Reinforcement learning
• Linear programming
• System identification
• Task analysis
• Dynamical systems
• Model Predictive Control
• Data-driven Model Predictive Control
• Accuracy Of Model
• Objective Function
• Dynamical
• Specific Tasks
• Control Synthesis
• Data-driven Control
• Neural Network
• Training Set
• Horizon
• Dynamic Model
• Performance Of Method
• Alternative Models
• Nonlinear Systems
• Multilayer Perceptron
• Hyperparameter Tuning
• Gaussian Process
• Bayesian Optimization
• Linear Quadratic Regulator
• True Dynamics
• Control Constraints
• Closed-loop Performance
• Design For Nonlinear Systems
• True Performance
• Hyperparameter Space
• Hold-out Set
• State Constraints
• Trajectory Optimization