Learn Before
Essay

Explain how to address an information gap in an autonomous driving pipeline.

Question: Suppose an autonomous driving pipeline consisting of car detection, pedestrian detection, and path planning performs far below humans, even though each component performs near human-level given its inputs. Explain why this happens, how it can be resolved without making path planning consume raw images, and the role of the additional detector.

Sample answer: This situation indicates that the pipeline is suffering from an information gap, meaning it is missing key information necessary for the overall task. While each component performs well individually on its given inputs, the path planning component lacks critical road information, such as lane markings, to make safe decisions. To resolve this without making the path planner process raw images directly, we can add a lane-marking detector to the pipeline. The lane-marking detector processes camera images and provides structured lane coordinates directly to the path planning component, supplying the missing information while keeping the path planner's inputs clean and structured.

Key points:

  • The overall system underperforms due to an information gap (missing information).
  • Each component is near human-level given its inputs, but the inputs to the path planner are insufficient.
  • Adding a lane-marking detector provides the missing lane information.
  • The lane-marking detector supplies this information without requiring the path planner to process raw camera images directly.

Rubric: The response must explain the concept of an information gap where individual components perform well but the overall system lacks necessary inputs. It must identify adding a lane-marking detector as the solution and state that this detector provides structured information to the path planner rather than forcing the path planner to consume raw camera images directly.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

References


Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related