01476770

Component

This research proposes and implements a reinforcement learning (RL) based signal control using R-Markov Average Reward Technique (RMART) accounting for neighborhood congestion information sharing. Results show significant improvement in system performance over both traditional fixed signal timing plans and real time adaptive signal control schemes. The comparison with reinforcement learning algorithms like Q-learning and SARSA algorithms indicate that RMART performs better at higher congestion level. Further, a multi-reward structure is proposed that dynamically adjusts the reward function with varying congestion state at the intersection. Finally, the empirical results also indicate that neighborhood information sharing improves the performance of reinforcement learning signal control algorithms in most cases.

This paper was sponsored by TRB committee AHB25 Traffic Signal Systems.

01470560

13-3227

English

34p

2013

Transportation Research Board 92nd Annual Meeting

Digital/other

Figures; References; Tables

Adaptive control; Information dissemination; Intersections; Neighborhoods; Real time control; Traffic signal control systems; Traffic signal timing

Reinforcement learning

Highways; Operations and Traffic Management; I72: Traffic and Transport Planning; I73: Traffic Control

Transportation Research Board Annual Meeting 2013 Paper #13-3227

TRIS, TRB, ATRI

Feb 5 2013 12:39PM