The thesis work will include various fields such as machine learning, simulation and optimization. Personal interest in programming is seen as benefit. The preferred language for coding is Java, but some other languages can be used if motivated.
Description of thesis work
To make a fleet of autonomous vehicles cost effective they need to act as a team. The reason is that decision of a single vehicle will more or less affect the operation of other vehicles.
This thesis focus on better understanding the machine learning technique Dyna2. Dyna2 combines learning and planning and has, for example, been succesfully been applied by Deep mind to the very challenging game board GO.
The double pong game is relevant to control of autonomous fleets because it is about coordinating objects. In the pong game, the controlled objects are two rackets. The game ends if the ball falls into the red area in the bottom of the screen. At all walls the ball will bounce. The objective is to move the padels so the game continous as for long time as possible.
By search it should be possible to identify the adequate short term actions of the objects. In the figure above: move the left padel to a more right position. The adequate horizon of the search is an open question.
Learning is about remembering good and bad states of the objets. In the game of double pong it is probably a bad state to let both rackets be at the same positions.
This master thesis handles the following questions:
- What are the adequate input signals to the artifical double pong player? An example signal is the ball x-position..
- How shall double pong physics be modelled?
- How shall bad states and actions be differentiated?
- How could the search per performed?
- How can the learning be performed? One can for example thing about training a feed forward neural network.
- How can differente problem approaches be compared? Compuational burden might be one aspect. A search horizon of 10 is more demanding compared to a horizon of 1.
The thesis work will include various fields such as machine learning, simulation and optimization. Personal interest in programming is seen as benefit. The prefered language for coding is Java, but some other languages can be used if motivated.
Thesis Level: Master and/or Bachelor
Starting date: Nov 20201 - Jan 2022
Number of students: 1-2
Tutor: Jonas, Dr., 0739-024761
The Volvo Group drives prosperity through transport solutions, offering trucks, buses, construction equipment, power solutions for marine and industrial applications, financing and services that increase our customers’ uptime and productivity. Founded in 1927, the Volvo Group is committed to shaping the future landscape of sustainable transport and infrastructure solutions. Countless career opportunities are offered across the group’s leading brands and entities that share a culture of Trust, Passion, High Performance, Change and Customer Success.
Volvo Autonomous Solutions constitute a new business area as of January 1, 2020, entering new and exciting territories for Volvo Group. We accelerate the development, commercialization and sales of autonomous transport solutions, focusing on defined segments for the on- and off-road space. The combination of strong tech expertise and skilled customer solutions creates innovative transport offers never seen before. We are constantly pushing our own skills and ability to drive change in a traditional industry to meet a growing customer demand. We are now looking for innovative, committed individuals to join us in our endeavor to create customer solutions that enhance safety, flexibility and productivity.
We want to get to know you
The journey begins! An email confirmation will be sent as soon as you submit your application. After this, it is still possible to update your personal profile by login in to your account. The hiring team will review your application together with the hiring manager. Shortlisted candidates will be contacted with information about the following steps.