Swiss sensor fusion specialist Baselabs has developed an algorithm that generates a consistent environment model from high-resolution raw sensor data.
Dynamic Grid accelerates the development of data fusion systems for automated driving functions, especially in challenging urban environments. This avoids time-consuming algorithm training so that automotive developers can develop driver assistance systems such as parking functions or traffic jam pilots with better performance than traditional tracking and grid methods.
Automated driving functions for urban areas set exceptionally high requirements on the environment model used. With high-resolution sensors to acquire the required data with a sufficient level of detail traditional algorithmic methods of sensor fusion are limited. The dynamic algorithm developed by Baselabs can process the high-resolution sensor data from radar or laser scanners at the raw data level. It is also possible to use cameras with semantic segmentation. This provides a self-consistent environment model that detects dynamic and static objects in the vehicle environment with high accuracy and robustness. In addition, it estimates free space to identify drivable areas or parking spaces. The algorithm runs on automotive CPUs in real-time and is implemented according to the ISO26262 safety process.
The Dynamic Grid algorithm is particularly suitable for driving functions for automation level 2 and above, including highly automated driving. Typical application areas are automated parking functions such as trained or valet parking, emergency braking functions with automatic avoidance, or traffic jam pilots a well as radar subsystems.
"With Dynamic Grid, we present a superior alternative to the combined use of traditional tracking methods and a static occupancy grid. By processing the data in an integrated manner in a self-contained algorithm, we avoid inconsistencies that the combination of two different methods in the traditional approach often entails. Dynamic Grid can show its strengths especially in scenarios with many objects and different directions of motion in the vehicle's environment. In addition, the algorithm can detect and track objects of any shape without extensive training," said