Co-Optimization of On-Ramp Merging and Plug-In Hybrid Electric Vehicle Power Split Using Deep Reinforcement Learning

Current research on Deep ReinforcementLearning (DRL) for automated on-ramp merging neglects vehicle powertrain and dynamics. This work considers a power-split Plug-In Hybrid Electric Vehicle (PHEV), the 2015 Toyota Prius Plug-In, using DRL. The control PHEV energy management are co-optimized such that DRL policy directly outputs power split between engine electric motor. testing results show can be successfully used co-optimization, leading to collision-free merging. When compared with sequential approaches wherein upper-level lower-level performed independently in sequence, we found co-optimization economic but jerky while may result collisions due neglecting limit constraints designing controller.