Case context: You are developing a self-driving car using a pipeline with three components: a car detector, a pedestrian detector, and a path planner. Evaluation shows that the car detector and pedestrian detector perform at near human-level. The path planner also performs near human-level when given ideal inputs. However, when the entire pipeline is tested on camera images, the vehicle's driving is far below human-level, often drifting out of lanes because it doesn't know where the lanes are.

Question: Based on the provided autonomous driving pipeline context, diagnose the primary system failure, explain why the overall system is underperforming despite strong individual components, and decide on a targeted pipeline modification to resolve the issue.

Sample answer: The primary system failure is an information gap; the pipeline is missing crucial information needed for proper path planning (specifically, lane locations). Even though the car and pedestrian detectors and path planner are near human-level given their inputs, the path planner does not receive lane-marking information. To resolve this, a lane-marking detector should be added to the pipeline to process camera images and feed lane coordinates directly to the path planner. This addresses the missing information without forcing the path planner to process raw camera images.

Key points:

• Diagnose the failure as an information gap (missing lane/road information).
• Explain that the path planner performs well given its inputs, but the inputs are insufficient for safe driving.
• Decides on adding a lane-marking detector to the pipeline to detect lane markings.
• Avoids making the path planner directly consume raw images by using the detector's output.

Rubric: The response must correctly diagnose the issue as an information gap (specifically missing lane/road structure information), explain that near-human component performance on limited inputs doesn't prevent failure if essential inputs are missing, and decide on adding a lane-marking detector to feed information to the path planner rather than sending raw images directly to the path planner.