CADET: Open-source platform benchmarks distributed cooperative AV autonomy

Researchers from UCLA have open-sourced CADET (Cooperative Autonomy through Distributed Experimentation Toolkit), a modular platform designed to systematically evaluate distributed cooperative autonomy in connected autonomous vehicles (AVs). Traditional AV pipelines rely on monolithic onboard computers for perception, planning, and control, but emerging cooperative autonomy leverages V2X (vehicle-to-everything) connectivity—including roadside units, edge servers, and cloud intelligence. CADET decouples the AV stack into composable modules that can be flexibly deployed across vehicles, infrastructure, and edge/cloud tiers, integrating state-of-the-art models with trace-driven network and workload emulation.

Through V2V and V2I experiments, CADET reveals that distributed deployment choices fundamentally shape safety: V2V intent packets outperform cloud-based perception, and RSU-assisted perception maintains safety until overloaded by concurrent requests. The framework provides synchronized model-, system-, and task-level instrumentation, supporting dataset-driven experimentation for systems and ML researchers to benchmark distributed inference workloads independently of full vehicle simulation. As large foundation models drive cloud deployment needs, CADET offers a critical tool for testing real-time decision-making under realistic network latency, compute heterogeneity, and multi-tenant contention.