Realistic Simulation of Vehicles for safer, more robust and cheaper development of autonomous vehicles

Important results from the project

The purpose of this project is to develop knowledge tools and processes that lead to better generation of synthetic training data for self-driving cars. Among these, measurable goals include examining whether we can remove 90% of the annotated data and maintain the same performance. During the project, an iterative approach has been used to achieve measurable goals. At the end of the project, the aforementioned goals have been achieved in more and more individual classes after each iteration, suggesting promising opportunities for the technology.

Expected long term effects

In the project, all practical sub-goals are met, resulting in software, knowledge and methods that all lead to the generation of useful synthetic data. Said knowledge, software and methods will be further developed with commercialization as the goal. In addition to future commercialization, results will also be used in another Vinnova FFI project that has already begun.

Approach and implementation

The first half of the project is about laying the foundations for those tools and more that are needed to perform experiments. The second half of the project has focused on an iterative process where the focus has been on improving, evaluating and learning from the pipeline that has been developed. The last quarter of the project went almost exclusively to final tests in order to answer research questions.