Towards Interactive Autonomous Vehicle Testing: Vehicle-Under-Test-Centered Traffic Simulation
Yiru Liu, Xiaocong Zhao, Jian Sun
TL;DR
The paper tackles the challenge of safe autonomous-vehicle testing by creating VUT-Centered environmental Dynamics Inference (VCDI), a Transformer-based framework that makes background traffic react to the Vehicle Under Test (VUT) through a VUT-centered, conditional trajectory inference approach. It introduces a Gaussian distributional cost function to model uncertainty and enable diverse, explainable scenario evolution, while integrating traditional trajectory inference with VUT anticipation to produce reactive, realistic background traffic. Key contributions include: (i) a VUT-centered interactivity framework, (ii) a learnable distributional cost for controllable scenario generation, and (iii) empirical validation on Waymo Open Motion data showing improved trajectory accuracy and richer scenario diversity, with code available. Overall, VCDI enhances AV testing by providing interactive, diverse, and explainable background traffic that reflects real-world uncertainty, enabling more robust evaluation of autonomous vehicles.
Abstract
The simulation-based testing is essential for safely implementing autonomous vehicles (AV) on roads, necessitating simulated traffic environments that dynamically interact with the Vehicle Under Test (VUT). This study introduces a VUT-Centered environmental Dynamics Inference (VCDI) model for realistic, interactive, and diverse background traffic simulation. Serving the purpose of AV testing, VCDI employs Transformer-based modules in a conditional trajectory inference framework to simulate VUT-centered driving interaction events. First, the VUT future motion is taken as an augmented model input to bridge the action dependence between VUT and background objects. Second, to enrich the scenario diversity, a Gaussian-distributional cost function module is designed to capture the uncertainty of the VUT's strategy, triggering various scenario evolution. Experimental results validate VCDI's trajectory-level simulation precision which outperforms the state-of-the-art trajectory prediction work. The flexibility of the distributional cost function allows VCDI to provide diverse-yet-realistic scenarios for AV testing. We demonstrate such capability by modifying the anticipation to the VUT's cost-based strategy and thus achieve multiple testing scenarios with explainable background traffic evolution. Codes are available at https://github.com/YNYSNL/VCDI.