AutoTrust: Benchmarking Trustworthiness in Large Vision Language Models for Autonomous Driving