Off-Policy Reinforcement based on a Safe Model Eco-Driving Education for Fully-Automated, Connected Hybrid Vehicles

Eco-driving reduces gas use. Eco-driving improves driving behavior and fuel economy without hardware modifications. Eco-driving strategies exist. Speed control is difficult in many driving conditions. This research develops an eco-driving approach to reduce car emissions in a variety of scenarios. DRL has helped eco-driving reduce fuel usage and travel time. Data-driven reinforcement learning can improve autonomous electric car speed planning. Eco-driving strategy reinforcement learning prototype. Three enhancements are offered. Off-policy learning and physics-based models improve sample efficiency. Second, constraint fulfillment can be achieved without externally rewarding proper behavior during training. Third, deep generative techniques to mimic a safe set ensure the route is possible. The reinforcement learning method can optimize car speed based on highway gradient and vehicle distance. Reinforcement learning outperforms evolutionary algorithms nearly perfectly.