Signal timing for LCV trucks on a road network using reinforcement learning

Abstract

Freight activity in urban networks is rising, and jurisdictions such as Ontario are encouraging the use of Long Combination Vehicles (LCVs) to consolidate freight loads. This thesis quantifies the delays and queueing on 16 intersections in the Region of Peel and introduces an adaptive signal-control strategy. Tested scenarios include (1) existing signal timing plans without LCVs and (2) with LCVs, (3) a single-intersection double deep q-network (DDQN) controller without LCVs and (4) with LCVs.

Introducing LCVs under existing signal timings raised network-wide delay by 14 % for all vehicles and 22 % for trucks when LCVs comprised just 1.7 % of traffic. The proposed DDQN was found to reduce average delays for all vehicles and trucks based on various conditions. Future work should extend the single intersection approach to a multi-agent framework and explore continuous-time action spaces for even finer control.