Learn Before
Why a Modular Pipeline Can Beat End-to-End Learning for Drone Landing
Question: In an autonomous drone landing system, explain why a modular design may be preferred over an end-to-end model. What factor most strongly drives this choice?
Sample answer: The main reason is data availability. An end-to-end landing policy would require large amounts of labeled examples that map raw sensor inputs directly to control outputs, but those examples are expensive and limited. A modular approach that separates perception, state estimation, and control can use the data that is actually available today, so it is more promising until much more end-to-end training data can be collected.
Key points:
- End-to-end methods need a great deal of direct input-to-output data.
- The available data often fits a modular system better.
- With current data limits, the modular approach is more practical and promising.
Rubric: The response should name data availability as the key factor and explain that the modular architecture better matches the data currently on hand.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Why is a modular driving system often considered a better near-term option than a fully end-to-end one?
True or False: A single end-to-end model is always the best choice for every machine learning problem.
Until more _____ is collected, a modular driving system is usually the better choice.
Match each machine learning strategy term to its description.
Put the argument for using a non-end-to-end design in autonomous driving in order.
Which task is presented as a strong success case for end-to-end learning?
True or False: A modular self-driving system can be a better fit when full end-to-end training data is limited.
An end-to-end driving model can map _____ directly to steering commands.
Match each data situation to the architecture choice it supports in self-driving research.
Order the steps for choosing between a full end-to-end model and a modular pipeline for a new machine learning problem.
Why a Modular Pipeline Can Beat End-to-End Learning for Drone Landing
Choosing the System Design for an Autonomous Delivery Cart
What makes end-to-end driving more practical?