Essay

Why separating perception from planning can simplify a robotics pipeline

Question: A robotics team is deciding between two designs for navigation. One option sends raw camera frames directly into the motion-planning module. The other option first uses a vision component to detect hallway boundaries and then passes that higher-level result to the planner. Explain the trade-off between these designs, focusing on how each choice affects module simplicity and implementation difficulty.

Sample answer: Sending raw camera frames straight to the motion-planning module gives that module too much to do. It would have to interpret unprocessed visual input and also decide where to move, which makes the task much harder to design and train. Using a separate vision component to extract hallway boundaries keeps the planner’s job smaller and more focused. The planner receives the information it needs without becoming responsible for low-level image interpretation, so the pipeline is easier to build and each part is simpler to train.

Key points:

  • Directly using raw camera frames makes the planning module handle an unnecessarily difficult problem
  • Raw-image input combines perception and planning inside one module
  • A separate vision component can supply the missing boundary information
  • Splitting the work keeps each module narrower in scope
  • Simpler modules are generally easier to implement and train

Rubric: Full credit requires identifying that direct raw-image input makes the planning module overly complex and that an intermediate perception step keeps the modules simpler and easier to build or train. Partial credit is appropriate if only one side of the trade-off is explained.

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