Abstract:In this paper, a novel approach is presented for modeling the interaction dynamics between an ego car and a bicycle in a traffic scenario using a hybrid reinforcement learning framework combined with a social value orientation (SVO) model. The proposed framework leverages the SARSA algorithm to learn the optimal policy for the ego vehicle while incorporating risk cost as the negative log-likelihood of collision. Additionally, a customized SVO model is introduced to capture the social preferences of the ego car and the bicycle, defining the SVO of each agent as a continuous variable between egoistic and cooperative orientations. Furthermore, a weight parameter is incorporated in the framework to regulate the influence of the SVO model on the learning process. We demonstrate the effectiveness of our approach through extensive simulations, showing that the ego car can balance between maximizing its reward and avoiding collisions while considering the social preferences of the agents. The obtained results are compared to other models in the literature, and it is shown that the proposed method contributes to the development of safe and efficient autonomous driving systems that interact with human-driven vehicles in a socially intelligent manner Note to Practitioners—This proposed framework is motivated by the pressing challenge of navigation for autonomous cars in complex urban driving scenarios and mixed traffic situations. With the increasing prevalence of autonomous vehicles on roads, developing intelligent navigation systems that can effectively interact with other road users has become essential. Our novel framework addresses this need by leveraging the SARSA algorithm to learn the optimal policy for the ego vehicle while incorporating risk cost as the negative log-likelihood of collision. Additionally, a customized SVO model is introduced to capture the social preferences of the ego car and the bicycle, defining the SVO of each agent as a continuous variable between egoistic and cooperative orientations. This enables autonomous vehicles to make informed decisions and navigate safely and efficiently. Our framework can enormously help the field of autonomous vehicle navigation and contribute significantly to developing safe, human-centric, and reliable transportation systems. The versatility of our approach is evident in its potential to support a network of autonomous vehicles interacting with multiple road users, thereby enhancing scalability. By leveraging the power of machine learning, our solution provides a robust and adaptable approach that can handle the diverse and ever-changing conditions of urban driving scenarios.

Social Intelligence for Autonomous Vehicles Interacting With Vulnerable Road Users

Social behavior for autonomous vehicles

Societal Intelligence for Safer and Smarter Transportation

Cognitive Internet of Vulnerable Road Users in Traffic: Predictive Neural Modulations of Road Crossing Intention

Analyze Drivers' Intervention Behavior During Autonomous Driving -- A VR-incorporated Approach

How Do Human-Driven Vehicles Avoid Pedestrians in Interactive Environments? A Naturalistic Driving Study

Social Interactions for Autonomous Driving: A Review and Perspectives

Interaction-Aware Decision Making for Autonomous Vehicles

Socially Intelligent Reinforcement Learning for Optimal Automated Vehicle Control in Traffic Scenarios

Making intersections safer with I2V communication

Online monitoring for safe pedestrian-vehicle interactions

Autonomous Vehicles That Interact With Pedestrians: A Survey of Theory and Practice

Summarizing Vehicle Driving Decision-Making Methods on Vulnerable Road User Collision Avoidance

Towards Socially Responsive Autonomous Vehicles: A Reinforcement Learning Framework with Driving Priors and Coordination Awareness

Intent Communication between Autonomous Vehicles and Pedestrians

Pedestrian-Vehicle Interaction in Shared Space: Insights for Autonomous Vehicles

How Shall I Drive? Interaction Modeling and Motion Planning towards Empathetic and Socially-Graceful Driving

Collision Avoidance Predictive Motion Planning Based on Integrated Perception and V2V Communication

Interactive cognition in self-driving

Pedestrian Moving Patterns during Potential Conflicts with 110 On-Road Driving Vehicles

Prediction of Pedestrian Spatiotemporal Risk Levels for Intelligent Vehicles: A Data-driven Approach