Safety margins in complex environments

Important results from the project

The primary goal of learning how to interact naturally and safely with other road users was successfully achieved and documented in three papers. In the first paper, a functional architecture was proposed to handle the needs of a self-driving vehicle and it was shown how a system could be taught to perform maneuvers efficiently and safely. In the second and third papers, it was shown how the system could be taught to take margins like the surrounding traffic. Some of the intermediate goals were skipped due to architectural choices which reduced their importance.

Expected long term effects

The project resulted in 3 articles. The first article was published at the ITSC conference in 2019 and lead to a collaboration with Berkeley university. This collaboration resulted in 2 papers which have been presented at the BAIR/BDD 2019 fall workshop in California. A secondary result of this project was a better understanding of architectural decisions for the control algorithms. Not only has this work influenced decisions made in the Zenuity software stack, but the ideas have been circulated within the research community in Gothenburg and Berkeley.

Approach and implementation

At the start of the project, architecture requirements were formulated for a control algorithm. A simulator was then created to study, design and test the decision and control components of a chosen framework. This allowed the interaction behavior of other road users to be varied in a controlled way. Algorithms could then be developed to interact with their fellow digital road users and tested under controlled conditions. The development was performed primarily at Zenuity and Volvo. Chalmers supported the development with conceptual discussions and supervision.