Model Predictive Control-Based Value Estimation for Efficient Reinforcement Learning

Reinforcement learning (RL) suffers from limitations in real practices primarily due to the number of required interactions with virtual environments. It results in a challenging problem because we are implausible to obtain a local optimal strategy with only a few attempts for many learning methods. Hereby, we design an improved RL method based on model predictive control that models the environment through a data-driven approach. Based on the learned environment model, it performs multistep prediction to estimate the value function and optimize the policy. The method demonstrates higher learning efficiency, faster convergent speed of strategies tending to the local optimal value, and less sample capacity space required by experience replay buffers. Experimental results, both in classic databases and in a dynamic obstacle-avoidance scenario for an unmanned aerial vehicle, validate the proposed approaches.

Authors

• Qizhen Wu
• Kexin Liu
• Lei Chen

Keywords

• Predictive models
• Neural networks
• Trajectory
• Data models
• Computational modeling
• Training
• Optimization
• Reinforcement learning
• Predictive control
• Estimated Values
• Neural Network
• Deterministic
• Value Function
• Transition State
• State Space
• Unmanned Aerial Vehicles
• Optimal Policy
• Model Predictive Control
• Reward Function
• Transition Function
• Fewer Data
• Reinforcement Learning Methods
• Actor-critic
• Inaccurate Model
• Model-free Reinforcement Learning
• Utilization Of Agents
• Dynamic Obstacles
• State Transition Function
• Model-based Reinforcement Learning
• Markov Decision Process
• Deep Q-learning
• Path Planning
• Prediction Step
• Environmental Model
• Loss Function
• Decisive Step
• Environmental Transitions
• Learning Speed
• Generalization Error