A Deep Reinforcement Learning Framework for Eco-Driving in Connected and Automated Hybrid Electric Vehicles

Zhu, Zhaoxuan; Gupta, Shobhit; Gupta, Abhishek; Canova, Marcello · 2023 · OpenAlex-citations

DOI: 10.1109/tvt.2023.3318552

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the eco-driving problem for Connected and Automated Hybrid Electric Vehicles (CAVs), aiming to minimize fuel consumption and travel time by co-optimizing speed trajectories and powertrain control strategies. The research is motivated by the potential of CAVs to leverage look-ahead information from connectivity and advanced mapping to operate more efficiently than human-driven vehicles. Specifically, the study focuses on hybrid electric vehicles (HEVs) navigating urban and highway environments with signalized intersections, a scenario where traditional optimization methods often face intractable computational demands due to the high dimensionality of combining energy management with velocity planning. The authors formulate the eco-driving problem as a Partially Observable Markov Decision Process (POMDP) and solve it using a Deep Reinforcement Learning (DRL) framework, specifically the Proximal Policy Optimization (PPO) algorithm with Long Short-Term Memory (LSTM) networks. To train and evaluate the agent, a high-fidelity co-simulation environment was developed, integrating a physics-based quasi-static nonlinear model of a 48V P0 mild-hybrid powertrain with a microscopic traffic simulator (SUMO). The environment utilizes real-world map data from Columbus, Ohio, generating 10,000 random trips for training and 100 for testing. The DRL agent receives observations including vehicle dynamics, battery state-of-charge, and Signal Phase and Timing (SPaT) data from upcoming traffic lights, and outputs torque commands for the internal combustion engine, electric motor, and mechanical brakes. The performance of the proposed DRL controller was benchmarked against a baseline representing human driving behavior, a trajectory optimization algorithm, and a deterministic optimal solution. The results demonstrate that the DRL controller reduces fuel consumption by more than 17% compared to the human-driver baseline while maintaining comparable travel times. This efficiency gain is achieved by intelligently modulating vehicle velocity to pass intersections during green phases and optimizing power split between the engine and battery. Notably, the DRL approach outperformed the more computationally demanding trajectory optimization method, offering a superior policy with minimal onboard computational requirements. The significance of this work lies in demonstrating that modern DRL algorithms can effectively solve complex, high-dimensional eco-driving problems for HEVs in real-world driving conditions. By providing a train-offline, execute-online methodology, the framework enables efficient energy management and speed planning without the heavy computational burden associated with classical optimization techniques. This approach facilitates the practical implementation of eco-driving strategies in connected and automated vehicles, contributing to reduced fuel consumption and emissions in mixed urban and highway scenarios.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• eco driving

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model