Sky-Drive: A Distributed Multi-Agent Simulation Platform for Socially-Aware and Human-AI Collaborative Future Transportation

Sky-Drive is the first comprehensive simulation platform that tightly integrates virtual reality (VR), multi-agent interactions, and human-centered AI approaches for multi-agent traffic simulation and human-centered autonomous agent research. Distinct from existing platforms, Sky-Drive introduces several key innovations: (a) a digital twin framework that creates high-fidelity virtual replicas of transportation systems for real-time monitoring and optimization; (b) a distributed multi-agent architecture that enables synchronized simulation across multiple terminals while maintaining precise real-time interactions between autonomous vehicles (AVs), human-driven vehicles (HDVs), and pedestrians; (c) a multi-modal human-in-the-loop framework that captures rich human behavioral data through various sensors, including steering wheels, eye-tracking cameras, and smartwatch sensors; (d) integration of fundamental models for enhanced human-machine collaboration and personalized decision-making such as large language models (LLMs) and vision language models (VLMs); (e) a novel human-AI bi-directional mechanism that facilitates effective knowledge exchange between human drivers and AI-enabled autonomous systems through both human feedback and domain knowledge from transportation science; (f) a hardware-in-the-loop module through ROS compatibility that enables direct verification of autonomous driving algorithms on physical platforms. Sky-Drive enables comprehensive research across various applications including VR-enabled vulnerable road user (VRU)-AV interactions, reinforcement learning-enabled autonomous driving policy learning, customized long-tail scenario generation, LLM-enabled personalized driving. Sky-Drive provides a unique environment for accelerating the development of safe and efficient AVs, while laying the groundwork for next-generation human-AI collaborative transportation systems.