Inverse Reinforcement Learning for Trajectory Imitation Using Static Output Feedback Control.

Area of Science:

• Control Theory
• Machine Learning
• Robotics

Background:

• Trajectory imitation is crucial for control systems, especially in dynamic environments with disturbances.
• Existing methods often require precise system models or extensive training data.
• Data-driven approaches offer a promising alternative for learning control policies.

Purpose of the Study:

• To develop a data-driven static output feedback (OPFB) control strategy for trajectory imitation in linear systems.
• To utilize inverse reinforcement learning (RL) to enable a learner system to mimic an expert's optimal trajectory.
• To analyze the stability, convergence, optimality, and robustness of the proposed algorithms.

Main Methods:

• An Expert-Learner framework is employed, where the learner reconstructs the expert's value function weights from input-output data.
• Three static OPFB inverse RL algorithms are proposed: a model-based baseline, a data-driven method using input-state data, and a data-driven method using only input-output data.
• The algorithms leverage measured data to infer and replicate the expert's control policy.

Main Results:

• The proposed data-driven inverse RL algorithms successfully enable trajectory imitation.
• The algorithms demonstrate effectiveness using only input and output data, reducing data requirements.
• Theoretical analysis confirms the stability, convergence, optimality, and robustness of the developed control strategies.

Conclusions:

• The study presents a viable data-driven approach for trajectory imitation control using static output feedback and inverse RL.
• The proposed methods offer a practical solution for systems where expert data is available but system models are unknown or complex.
• Simulation results validate the effectiveness of the algorithms in achieving accurate trajectory imitation under external disturbances.