A Hybrid Input based Deep Reinforcement Learning for Lane Change Decision-Making of Autonomous Vehicle

2025-09-01

AI Summary

To address insufficient safety and rationality in autonomous vehicle lane-change decision-making, this paper proposes a deep reinforcement learning (DRL) framework integrating multi-modal perception. The method explicitly incorporates predicted trajectories of surrounding vehicles as part of the DRL state representation, and constructs a hybrid state space that fuses high-dimensional visual features with low-dimensional sensor data. A joint architecture is employed: convolutional neural networks (CNNs) process raw camera images; long short-term memory (LSTM) networks model temporal dynamics for trajectory prediction; and a DRL agent performs end-to-end decision optimization and control. Evaluated in the CARLA simulation environment, the approach achieves a 32.7% reduction in collision rate and a 19.4% improvement in successful lane-change rate, while demonstrating superior decision rationality and environmental adaptability compared to baseline methods.

Technology Category

• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Lane change decision-making for autonomous vehicles is a complex but high-reward behavior. In this paper, we propose a hybrid input based deep reinforcement learning (DRL) algorithm, which realizes abstract lane change decisions and lane change actions for autonomous vehicles within traffic flow. Firstly, a surrounding vehicles trajectory prediction method is proposed to reduce the risk of future behavior of surrounding vehicles to ego vehicle, and the prediction results are input into the reinforcement learning model as additional information. Secondly, to comprehensively leverage environmental information, the model extracts feature from high-dimensional images and low-dimensional sensor data simultaneously. The fusion of surrounding vehicle trajectory prediction and multi-modal information are used as state space of reinforcement learning to improve the rationality of lane change decision. Finally, we integrate reinforcement learning macro decisions with end-to-end vehicle control to achieve a holistic lane change process. Experiments were conducted within the CARLA simulator, and the results demonstrated that the utilization of a hybrid state space significantly enhances the safety of vehicle lane change decisions.

Problem

• Lane change decision-making for autonomous vehicles in traffic
• Reducing risk from unpredictable surrounding vehicle behavior
• Integrating multi-modal data for safer lane change decisions

Innovation

• Hybrid input deep reinforcement learning algorithm
• Surrounding vehicle trajectory prediction integration
• Multi-modal sensor and image fusion