Learning to Collide: An Adaptive Safety-Critical Scenarios Generating Method
This addresses the long-tail and rare event challenges in autonomous driving safety evaluation, though it is incremental in improving scenario generation efficiency.
The paper tackles the problem of evaluating autonomous driving algorithms in safety-critical scenarios by proposing a generative framework that creates diverse and risky scenarios more efficiently than grid search or human design methods, as demonstrated in simulation experiments.
Long-tail and rare event problems become crucial when autonomous driving algorithms are applied in the real world. For the purpose of evaluating systems in challenging settings, we propose a generative framework to create safety-critical scenarios for evaluating specific task algorithms. We first represent the traffic scenarios with a series of autoregressive building blocks and generate diverse scenarios by sampling from the joint distribution of these blocks. We then train the generative model as an agent (or a generator) to investigate the risky distribution parameters for a given driving algorithm being evaluated. We regard the task algorithm as an environment (or a discriminator) that returns a reward to the agent when a risky scenario is generated. Through the experiments conducted on several scenarios in the simulation, we demonstrate that the proposed framework generates safety-critical scenarios more efficiently than grid search or human design methods. Another advantage of this method is its adaptiveness to the routes and parameters.