Scenario tagging

Content / Page
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 and training these models requires a huge amount of annotated data. Annotating data is time consuming and our teams at Kognic have developed an annotation platform to make the annotation workflow efficient, including all aspects from UI/UX to integration of interactive deep learning algorithms.

In order to provide additional insights to our clients and ease the work of annotators, Kognic is looking for innovative ways to develop scene understanding.

Problem description

In order to adapt to the current scenario, a self-driving vehicle must be aware of his surroundings, but also be able to understand the context and the situation of infrastructure and other road users. This global understanding is fundamental to help the car adapt its speed, direction and global behaviours on the road.

To understand if the AV has "seen" a specific scenario, we want to create an algorithm able to find sequences that contain a specific scenario.

The list of scenarios is not set in stone, a part of the thesis will be to define the scope and understand what scenarios are realistic based on the data provided. Some possible examples could be described as follows:

Close cut-in

Front vehicle braking suddenly

Vulnerable Road User on the ego road

There is also the draft regulation proposal from EU for AV’s which contains a list of scenarios that should be passed, this could be a good set of scenarios.

We expect the students to:

Perform a thorough literature study and select one or two suitable situations that is useful for a scenario classification function

Participate in the selection and preprocessing of required data necessary for training

Implement and validate the selected solutions

Suggest an effective way to implement the developed solution to Kognics' platform

Goals & Challenges

The goal is to develop one or a set of machine learning algorithms that can take annotated data and find if and when the scenario is happening within a sequence. 

Applicant profile

We seek two students with an interest in computer vision and deep learning (CNN, RNN). We also expect the students to have experience with Python and preferably Pytorch framework.

Project information

Project start: 16th Jan 2023

Duration: 20 weeks

Last date of application: 5th of Dec


The remuneration for this internship will be 1000 SEK/week (before taxes), which will be paid in whole when the thesis is completed and we have received a copy. 

The total remuneration will be 20000SEK for the project per student.

Contact information:

Andreas Falkovén:
Tommy Johansson:


Master Thesis Proposals