Short Answer

Why modular driving systems are easier to train than fully end-to-end ones

Question: What data-collection burden makes a fully end-to-end car control model harder to build than separate perception modules?

Sample answer: A fully end-to-end driving model needs many examples that pair camera views with the exact control action a human driver took, such as steering, braking, or accelerating. Gathering those matched examples is costly and slow. By comparison, a perception module can often be trained with ordinary object labels, like marking traffic lights, buses, or lane markings in images, and those labels are easier to obtain.

Key points:

  • End-to-end models depend on paired scene-and-control records, which are expensive and time-consuming to collect.
  • Smaller perception modules can use standard image labels for objects or road features, which are simpler to gather.

Rubric: The answer must state that end-to-end training needs large numbers of camera frames matched to driving actions, making data collection difficult and costly, while intermediate perception tasks can rely on easier-to-collect labeled images.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI

Related