Improving a Proportional Integral Controller with Reinforcement Learning on a Throttle Valve Benchmark
Abstract
This paper presents a learning-based controlstrategy for non-linear throttle valves with an asymmetrichysteresis leading to a near-optimal controller. We start witha carefully tuned Proportional Integrator (PI) controller andexploit the recent advances in Reinforcement Learning (RL)with Guides to improve the closed-loop behavior by learningfrom the additional interactions with the valve. We test theproposed control method in various scenarios on a throttlevalve benchmark, highlighting the benefits of combining bothPI and RL frameworks to improve control performance in non-linear stochastic systems. In all the experimental test cases, theresulting agent has a better sample efficiency than traditionalRL agents and outperforms the PI controller.
| Origin | Files produced by the author(s) |
|---|