Autonomous vehicles learn to find the best path

Kristoffer Bergman, doctoral student in the Division for Automatic Control, describes in his licentiate thesis methods to enable autonomous vehicles find the fastest or most energy-efficient path from a starting point to a destination, even if obstacles are in the way.

Kristoffer Bergman Photo credit David BrohedeGetting autonomous vehicles to travel along straight roads without obstacles is one thing, but is it also possible for them to autonomously manoeuvre past obstacles in order to deliver goods? Can a drone fly low without colliding with trees, chimneys, etc.? And how can the vehicle plan its route to make it as efficient as possible? These are questions that Kristoffer Bergman seeks to answer in his licentiate thesis.

Kristoffer Bergman is an industry-based doctoral student in the Division for Automatic Control, employed by Saab Dynamics and working in the large Wallenberg AI, Autonomous Systems and Software Program, WASP.

Optimal Control

“I’ve worked with problems that can be solved with a method in automatic control known as optimal control. It’s a powerful tool, whose applications include helping to send people to the moon. It does, however, have limitations when there are many obstacles to avoid”, says Kristoffer Bergman.

His research has aimed at developing reliable and robust methods that autonomous vehicles can use to plan their route around various obstacles.

“We also want to be able to optimise the route efficiently, such that it gives, for example, the quickest journey or the one lowest energy consumption”, he says.

WASP-licar; Per Boström-Rost (till vänster), Kristoffer Bergman (till höger)Kristoffer Bergman and Per Boström-Rost (to the left) Photo credit Magnus JohanssonThe thesis presents three scientific contributions to the field. The first uses optimal control, in which and large and complex problems are solved by first solving subsidiary problems of increasing difficulty.

The second contribution describes how to calculate a library of pre-programmed and optimised fundamental manoeuvres, known as “motion primitives”, that can subsequently be combined, depending on the situation. This has previously been time-consuming and required manual work, but Kristoffer Bergman proposes a framework that optimises and automates the calculations.

In combination

“Some motion primitives can be linked together online, and robust algorithms are available to do this. But there is a risk that the algorithms generate poor solutions on the basis of the limited set of precalculated motion primitives”, he explains.

The third contribution presented by the thesis, therefore, shows how to combine the method using motion primitives with a method based on optimal control.

“The combination means that the calculations take lightly longer. The solution, however, is significantly better, since there is now no longer a limitation to use the library of motion primitives.”

Kristoffer Bergman sees huge advantages in being a part of WASP.

“WASP has enabled me to gain a comprehensive view, and a good grip of the complete system. The courses and conferences we have had have given not only a breadth to our knowledge but also an impressively wide contact network”, he concludes.

On motion planning using numerical optimal control, Kristoffer Bergman, Division for Automatic Control, Department of Electrical Engineering, Linköping University 2019. Supervisor Daniel Axehill.

Translated by George Farrants

ContactShow/Hide content

ReseachShow/Hide content

Latest news from LiUShow/Hide content