eduzhai > Applied Sciences > Engineering >

An online evolving framework for advancing reinforcement-learning based automated vehicle control

  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: In this paper, an online evolving framework is proposed to detect and revisea controller s imperfect decision-making in advance. The framework consists ofthree modules: the evolving Finite State Machine (e-FSM), action-reviser, andcontroller modules. The e-FSM module evolves a stochastic model (e.g.,Discrete-Time Markov Chain) from scratch by determining new states andidentifying transition probabilities repeatedly. With the latest stochasticmodel and given criteria, the action-reviser module checks validity of thecontroller s chosen action by predicting future states. Then, if the chosenaction is not appropriate, another action is inspected and selected. In orderto show the advantage of the proposed framework, the Deep Deterministic PolicyGradient (DDPG) w and w o the online evolving framework are applied to controlan ego-vehicle in the car-following scenario where control criteria are set byspeed and safety. Experimental results show that inappropriate actions chosenby the DDPG controller are detected and revised appropriately through ourproposed framework, resulting in no control failures after a few iterations.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...