Scenario Orchestration for Closed-loop Policy Evaluation of Autonomous Driving
Scenario Orchestration for Closed-loop Policy Evaluation of Autonomous Driving
This research introduces a scenario orchestrator designed for evaluating autonomous driving policies. The system continuously monitors traffic states and adapts non-ego actors through re-timing, re-routing, and role reassignment to maintain intended scenario interactions during closed-loop testing. By solving constrained optimization problems at each step, the method preserves interaction intent while limiting interventions and ensuring traffic feasibility. Experiments in the CARLA simulator demonstrate that this approach realizes targeted interactions more reliably than scripted or learned actor-level baselines, particularly under varying initial conditions and ego behaviors. The work supports closed-loop scenario orchestration as a tool for controlled and reproducible policy evaluation.
The paper proposes a scenario orchestrator for autonomous driving that uses constrained optimization to adapt non-ego actors in real-time, distinguishing between scenario realization and policy failure to enable closed-loop evaluation.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
and maintaining traffic feasibility. We further distinguish scenario realization from policy failure, allowing a test to first be judged based on whether the intended interaction occurred and then by how the ego policy responded. Experiments in the CARLA simulator across multiple scenario families and driving policies show that orchestration realizes targeted interactions more reliably than scripted and learned actor
Open source S4
arXiv:2611.09888v1 Announce Type: new Abstract: Scenario-based simulation is central to evaluating autonomous-driving policies, but a scenario scripted before execution (i.e., with predetermined, fixed actor routes and speeds) may not preserve its intended interaction once the tested ego policy responds in closed loop. Scripted scenarios cannot adapt to unanticipated ego behavior, while learned traffic actors may rem
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org