Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments

TL;DR

A novel “adaptive visual sector” mechanism is introduced that mimics the dynamic allocation of attention human drivers exhibit based on factors like spatial orientation, proximity, and driving speed and develops a “dynamic traffic graph” using Convolutional Neural Networks and Graph Attention Networks to capture spatio-temporal dependencies among agents.

Abstract

In the burgeoning field of autonomous vehicles (AVs), trajectory prediction remains a formidable challenge, especially in mixed autonomy environments. Traditional approaches often rely on computational methods such as time-series analysis. Our research diverges significantly by adopting an interdisciplinary approach that integrates principles of human cognition and observational behavior into trajectory prediction models for AVs. We introduce a novel "adaptive visual sector" mechanism that mimics the dynamic allocation of attention human drivers exhibit based on factors like spatial orientation, proximity, and driving speed. Additionally, we develop a "dynamic traffic graph" using Convolutional Neural Networks (CNN) and Graph Attention Networks (GAT) to capture spatio-temporal dependencies among agents. Benchmark tests on the NGSIM, HighD, and MoCAD datasets reveal that our model (GAVA) outperforms state-of-the-art baselines by at least 15.2%, 19.4%, and 12.0%, respectively. Our findings underscore the potential of leveraging human cognition principles to enhance the proficiency and adaptability of trajectory prediction algorithms in AVs. The code for the proposed model is available at our Github.

Figures (3)

• Figure 1: Adaptive dynamic mechanism of the human driver's attention. Left: The human driver's visual field can be roughly divided into three parts: central, fringe, and peripheral vision fields. The central vision field receives the most attention, while visual information from the fringe and peripheral vision fields is prioritized during maneuvering changes, which are observed through side and rearview mirrors and receive comparatively less attention and observation time. Right: The coverage angle of the central vision field adjusts rapidly with velocity in real-time.
• Figure 2: A general overview of Gava. The original data is processed through the Context-Aware Module to extract high-dimensional temporal information, resulting in the formation of the context feature. Meanwhile, raw data is processed through the Visual-Aware and Interaction-Aware Modules, yielding Visual Features. Finally, the Priority-Aware Module is employed to generate multi-modal trajectory predictions based on the fusion of the two features.
• Figure 3: Visualization of Predictions for Complex Traffic Situations in NGSIM (above) and HighD (below) Datasets. The yellow rectangles denote the target vehicle, while other rectangles represent the neighboring vehicles.