Motivation and background
The development of self-driving vehicles is an important and challenging task. The vehicles' perception systems are often based on deep learning models and huge sets of annotated data is required to train them. Annotating this data is a time consuming task where annotators manually label the data according to a client’s needs.
One large challenge in this process is to instruct the annotators so that they can execute on the clients needs with sufficient quality and speed. The most common approach today is to use written instruction documents to explain how and at what detail to annotate the data. The instruction documents used today are usually written by engineers at the client side. This leads to a high variance in how the documents are written and structured, as well as a use of language that might not be optimal for its audience. We strongly believe that we can find new ways and improvements in how we communicate these instructions to our users.
The objective with this thesis will be to investigate and suggest how to communicate instructions to annotators in a user-friendly way that accelerates learning. As our annotation teams are spread throughout the world and have a varying degree of technical literacy in the field of machine learning and self-driving vehicles - we expect this perspective to be incorporated into the suggested solutions.
We believe you will achieve this by:
- Analyzing the current solution and its strengths and weaknesses in relation to learnability and other usability aspects.
- Perform a literature study around the topic of learning.
- Design multiple concepts that take your learnings into consideration.
- Test and evaluate the concepts with users.
- Select and iterate one or multiple concepts based on the evaluation insights.
- Suggest one or multiple solutions which you believe succeeds at communicating instructions in a way that facilitates learning.
The solution can be everything from a written guide on how to write instructions to a product concept.
Who are you?
You are two students who have a background in user-centered design. We expect you to have experience in using design processes - going from identifying a usability problem to providing a user-friendly solution.
We recommend that you submit your application using Teamtailor as soon as possible, as we interview continuously.
If you have any questions regarding the thesis project, please contact email@example.com.
Project start: 15th Jan 2022
Duration: 20 weeks
Last date of application: 5th of Dec
Annotell was founded in 2018 by Oscar Petersson and Daniel Langkilde, two engineering physicists working in the field of Deep Learning. Our mission is to make safe perception for autonomous mobility possible. We now support world-leading companies in the field of Autonomous Driving, Advanced Driving Assistance Systems and Active Safety development worldwide.
Our office is located at Lindholmen, just by the beautiful waterfront and Lindholmen Science Park.