Addressing perception robustness and inference latency in lead-vehicle ACC scenarios (October week 2)

Expected goals:

• Mitigating perception degradation from vehicle shadows and horizon lane-merging issues when a lead vehicle is present.
• Improving perception latency and throughput (FPS) for high-speed scenarios (e.g. 50 km/h) by exploring Bird’s-Eye-View (BEV) networks and optimizing image preprocessing pipelines.
• Investigating CARLA-specific retraining and tuning (higher resolution input, ROI adjustment) while evaluating alternatives or maintaining YOLOP.

More details in this link.