driveblocks Mapless Autonomy Platform receives major feature updates with new 2023.02 version

driveblocks, a software provider for commercial vehicle autonomy, releases the 2023.02 version of its Mapless Autonomy Platform. The platform includes algorithms for object detection, environment perception, sensor-fusion and trajectory planning in various commercial vehicle use-cases, such as highway driving, mining, container logistics or agricultural applications. It is designed to be hardware agnostic and supports various multi-modal sensor-setups with minimal parametrization and retraining efforts.


Major additions to the environment perception and sensor-fusion

 Coming from the 2023.01 release in January, the team has added several new features to the perception and sensor-fusion parts of the software platform:

  • Estimation of drivable space boundaries using visual features in camera images, such as pylons, vertical panels, metal barriers, gras, and gravel in construction sites and off-road scenarios

  • Support for complex line geometries within line detection networks such as overlapping lines and lane merging scenarios

  • A novel probabilistic ground estimator taking into account the neighborhood of the considered areas and improved results on non-flat surfaces

  • A novel fusion- and tracking algorithm for static and dynamic objects

 Additional minor improvements of the platform include:

  • Several fixes and additional information supported for Lanelet2 formatted output of environment and driving corridor model.

  • Decreased resource requirements for neural network to support a larger number of cameras on a single autonomous driving compute unit.

  • Full support for intensity channels in LIDAR perception stack

  • Increased stability of driving corridor estimation via several minor bugfixes

One of the major focus areas of the driveblocks Mapless Autonomy Platform is the generation of a consistent environment model, including drivable space, potential driving corridors as well as stationary and dynamic object positions and sizes. The upcoming releases will add several capabilities around these tasks and will continue to improve upon the range, reliability, and robustness of the algorithms.