CapCo: Automating Carla-Apollo Co-Simulation and Scenario Fuzzing

Testing industrial-grade autonomous driving systems (ADSs) such as Baidu Apollo demands scalable, high-fidelity simulation across diverse and high-risk traffic scenarios. From a software engineering perspective, this motivates reproducible and automatable test infrastructures that support regression testing at scale. However, current simulation toolchains either rely on deprecated platforms (e.g., LGSVL) or require extensive manual setup, hindering repeatability and fuzzing-based evaluation. We present CapCo, an open-source framework that automates co-simulation between Apollo and the Carla simulator, enabling closed-loop, scenario-based testing at scale. CapCo automates the full workflow - from launching Apollo and configuring scenarios to executing and monitoring tests - without manual intervention, and integrates Apollo into the Carla Leaderboard to ensure synchronized communication and real-time feedback. CapCo supports multiple fuzzing strategies and includes a curated set of high-risk traffic scenarios, with a user-friendly interface for configuring experiments, tracking progress, and analyzing results. Video: https://youtu.be/CUPHQfnTVss. Code: https://github.com/TayYim/Capco.