Case Study

Redesigning a Self-Driving Pipeline to Support Missing Lane Information

Case context: You are overseeing the development of a modular self-driving car pipeline. The path-planning module needs lane-marking boundaries to make decisions, but it currently does not receive this information. A team member suggests modifying the path-planning module to receive raw camera images directly so it can find the lane markings itself. They argue this is the fastest way to get the missing information to the planner.

Question: Evaluate the team member's proposal based on the task simplicity design principle. Propose an alternative pipeline design and explain why it is superior for building and training the modules.

Sample answer: The team member's proposal should be rejected because feeding raw camera images directly to path-planning violates task simplicity. It forces the path planner to process raw images, making the module extremely complex. A superior alternative is to insert an intermediate 'Detect lane markings' module. This module takes raw camera images, identifies the lane markings, and passes only this structured information to the path planner. This avoids making either the detector or the path-planner overly complex to build or train.

Key points:

  • Reject the proposal of routing raw camera images directly to the path-planning module.
  • Explain that direct routing violates task simplicity by complicating the planner's input and task.
  • Propose adding a 'Detect lane markings' component between the camera and the path planner.
  • State that the redesigned modular structure prevents modules from becoming overly complex to build and train.

Rubric: The response should reject the proposal because it violates task simplicity by having the planner process raw images. It must propose inserting an intermediate lane-marking detector and explain that this structure keeps individual modules simple to build and train.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related