Dynamic Decision-Making Framework for Autonomous Vehicles in Urban Environments with Strong Interactions

Abstract

Autonomous techniques are becoming increasingly integrated into our daily lives. Many advanced driver assistance systems (ADAS) functions, such as lane-keeping assist and car-following, are already implemented in manufactured vehicles. However, achieving true autonomy still poses many challenges. For instance, in urban areas with diverse types and numbers of traffic participants, the interactions are highly complex. Considering these strong interactions are time consuming and challenging. Additionally, the fast-changing nature of urban driving scenarios requires the decision-making of self-driving vehicles to be performed in real-time. The various behaviors of different traffic participants also make the corresponding decision-making challenging. Finally, in urban traffic scenarios, following traffic rules is the premise of any decision-making. However, the extensive and often difficult interpretation of traffic rules adds another layer of complexity.

This thesis aims to bring the decision-making process of autonomous driving techniques closer to real life by proposing a motion planning and decision-making framework for autonomous vehicle urban driving that addresses the aforementioned challenges. The framework utilizes game theory to formulate and consider strong interactions.

The behaviors of surrounding traffic participants are estimated more accurately by extracting realistic behavioral characteristics from real-world driving datasets. This helps establish more realistic modeling and estimation of various kinds of traffic participants, including aggressive, neutral, and conservative types. Accurate modeling of traffic participants improves the quality of interaction formulation, leading to sounder decision-making.

To ensure adherence to traffic regulations, the proposed framework extracts right-of-way information from traffic rules to generate behavioral parameters. This acts as a bridge integrating traffic rules into the decision-making process. The traffic rules not only help the ego vehicle estimate the future behaviors of surrounding traffic participants by extracting precedence but also generate rule-adhering behaviors for the ego vehicle.

Additionally, to improve the real-time performance of the framework in very crowded urban scenarios, the framework is equipped with a human-like attention-based traffic actor filter. This enables the autonomous vehicle to focus on critical traffic participants with a higher risk of collision, simplifying the decision-making and planning process, reducing computational effort, and ensuring real-time performance.

To implement the proposed framework in the real world, a full-size vehicle platform was developed, equipped with appropriate hardware sensors and onboard computers. A corresponding hierarchical software system was also developed to ensure the vehicle's operation.

The proposed framework was tested in both simulation and real-world scenarios. The results demonstrate that the autonomous vehicle can properly estimate the types of traffic participants by observing their behavior using the proposed technique. The vehicle then behaves according to these types, enabling interactive and human-like planning and decision-making at intersections. Furthermore, the autonomous vehicle is able to consider and adhere to traffic rules in very complicated urban traffic scenarios. These results demonstrate that the algorithm can make safe and efficient decisions in various urban traffic scenarios involving multiple types of traffic participants in real-time.

The simulation results show that the autonomous vehicle is able to properly estimate the types of traffic participants by observing their behavior using the proposed technique. Then the autonomous vehicle behave according to the types of those traffic participants to enable interactive and human-like planning and decision making at intersections.