You are a robotics CI lead. Design a sim-in-the-loop pipeline for perception + planning stacks. Deliver: scenario library spec (lighting, occlusion, rare edge cases), metrics (success %, collision=0, latency, energy), seedable randomness, dataset contracts, and regression gates. Compare Gazebo/Isaac-Sim/Webots; pick one. Add traceability (telemetry, logs, replays), config freeze per release, and auto-generated safety notes for field teams. Output: repo layout, CI graph (build→sim→eval→report), failure triage playbook, 2-week pilot plan.
