Q-Guided Stein Variational Model Predictive Control via RL-informed Policy Prior

Researchers have developed Q-SVMPC, a new Model Predictive Control method that combines reinforcement learning with Stein variational inference to improve trajectory optimization. The approach addresses limitations in existing MPC methods that often converge to single solutions, instead maintaining diverse solution paths for better performance in robotics applications.

• →Q-SVMPC combines Model Predictive Control with reinforcement learning and Stein variational inference to maintain diverse trajectory solutions.
• →The method addresses mode collapse issues in existing gradient-based and sampling-based MPC solvers.
• →Experimental results show improved sample efficiency, stability, and robustness compared to traditional MPC and model-free RL methods.
• →The approach was tested on navigation, robotic manipulation, and real-world fruit-picking tasks.
• →The research contributes to learning-based MPC by treating it as trajectory-level posterior inference problem.