A new reinforcement learning algorithm is introduced which can be applied over a continuous range of actions. Reinforcement learning benefits over other types of learning systems in requiring no dedicated training and evaluation phases of learning; instead, such systems progressively adapt to the given environment. The learning algorithm is reward-inaction based, with a set of probability density functions being used to determine the action set. The learning sub-system sends an action or set of actions to the environment which then returns a scalar value, via the performance evaluation function, indicating the quality of that action. The performance evaluation function encodes the explicit goals and objectives of the learning system and returns a scalar value, indicating the quality of the applied actions. An experimental study is presented, based on the control of a semi-active suspension system on a road-going, four wheeled, passenger vehicle. The control objective is to minimize the mean square acceleration of the vehicle body, thus improving the ride isolation qualities of the vehicle. This represents a difficult class of learning problems, owing to the stochastic nature of the road input disturbance together with unknown high order dynamics, sensor noise and the non-linear (semi-active) control actuators. The learning algorithm described here operates over a bounded continuous action set, is robust to high levels of noise and is ideally suited to operating in a parallel computing environment. A standard passenger vehicle fitted with continuously variable controllable semi-active dampers and wheel and body accelerometers was mounted on a servo-hydraulic four-poster road simulator. The learning methodology was applied to the problem of improving the ride comfort of the vehicle with a performance objective of reducing the vertical body accelerations. Suspension deflection, wheel velocity and body velocity at each corner of the vehicle were measured and used for feedback control.
Continuous action reinforcement learning applied to vehicle suspension control
Zeitkontinuierliche lernende Schwingungsregelung einer Radaufhängung
Mechatronics ; 7 , 3 ; 263-276
1997
14 Seiten, 10 Bilder, 1 Tabelle, 10 Quellen
Article (Journal)
English
Suspension control system based on reinforcement learning
European Patent Office | 2020
|Air spring applied to vehicle suspension, vehicle suspension and vehicle
European Patent Office | 2022
|VEHICLE HAVING SUSPENSION WITH CONTINUOUS DAMPING CONTROL
European Patent Office | 2017
|VEHICLE HAVING SUSPENSION WITH CONTINUOUS DAMPING CONTROL
European Patent Office | 2017
|