When cars can find their way around a new environment as well as humans can, they will be a lot closer to being truly autonomous vehicles. Help Veoneer make autonomous vehicles that are a safe and trustworthy transportation solution.

Autonomous Driving (AD) has the potential to transform many aspects of our lives, making goods cheaper and mobility easier for everyone, especially the most vulnerable and disadvantaged members of our society. We at Veoneer are excited to make this dream a reality.

Human drivers can navigate their way through very challenging road conditions. With only our eyes and brains, we can build a mental map of a road we have never seen before, even as we keep track of our own location within that road. By contrast, current AD solutions tend to require that a very precise High Definition map (HDMap) of the area be built up before attempting to drive there. This typically leaves a lot of locations (parking garages, private driveways) and situations (degraded or missing lane markers, roads with construction) beyond the reach of AD vehicles, and hampers or prevents many transformative applications (door-to-door transportation, fully autonomous delivery).

To truly enable fully autonomous driving, we need AD vehicles to have a self-localization and novel environment mapping capability comparable to our own. Modern solutions to the Simultaneous Localization and Mapping (SLAM) problem can provide such a capability. SLAM solutions can build a map by tracking “landmarks” over time and building up a map of the vehicle’s surroundings.

Many of the basic problems in SLAM have been solved to some degree, but AD is a safety critical application where lives are at stake. Additional problems remain to be solved to make SLAM accurate, robust, and reliable enough to safely guide autonomous vehicles. For example, a human might mistake one sign for a similar looking one, building an incorrect mental map as a result. When we realize our mistake, we quickly correct our mental map and continue on. A basic SLAM system, by contrast, has no way to remove the false association and its results quickly become grossly inaccurate.

In this project, you will learn about SLAM, build a functional baseline SLAM system from open source libraries, explore some of these additional challenges and experiment with solutions.

More information