Demonstrators

SCD 6.4: Localisation and 3D mapping

BUT

BUT, IMA

Demonstrator enables direct inter-vehicle sharing of 3D maps of surroundings. The information about vehicle neighbourhood is crucial for real-time sharing of obstacles among vehicles, thus improving the safety of operation and the optimality of driving process. However, since the data-flow from sensors sensing around the vehicle is extremely high, it is important to reach significant bitrate reduction to be able to share obstacles in real-time. Such a bitrate reduction is performed using AI-based feature extraction, segmentation of clustering, followed by intelligent inter-agent data sharing, when only the simplified and necessary information are extracted from the raw sensor and shared; instead of sharing all the data. Another aim is to protect the vehicle body itself and the humans around it when the vehicle is stopped. The mapping of the nearest area of the vehicle will protect from damages and injuries, primarily during the door opening.

The greatest benefit of the mentioned technology is the ability to share obstacles between autonomous vehicles in real-time. Vehicles equipped with this system can "see around corners," gaining awareness of obstacles even before they appear in the vehicle's field of view. This allows them to react promptly, avoiding collisions or optimizing their movement. Another advantage is the increased robustness of obstacle detection and classification, based on the principle of "two heads are better than one." Information about a single obstacle is received by a vehicle from multiple surrounding vehicles simultaneously. Since each sender observes the obstacle from a different angle, the data fusion of these inputs enables the elimination of potential errors in detection or classification.

Example of a local map containing parked vehicles (blue) in combination with raw lidar data (green).