Selfdriving Software Stack
SafeAD offers a state of the art full selfdriving stack. We license the full stack or individual modules to our customers.
Our perception uses deep neural networks to detect objects in 3D from a single or multiple cameras using learned depth estimation. Objects can be tracked by incorporating temporal information. Additionally we offer classical bounding box detections and semantic segmentation.
Our visual localization from multiple cameras is fused with odometry data from wheel encoders and measurements from innertial sensors. This allows fast, robust and centimeter accurate localization. Our multi-camera setup allows to stay localized even when large parts of a camera are occluded by other vehicles or dirt while the fusion of odometry allows for high update rates of more than 100Hz.
Planning & Control
We offer a cost-based planner which guarantees safety by applying hard constraints to obstacles and lane boundaries. Further, soft constraints allow for a comfortable and smooth ride.
The trajectory is passed to an MPC controller which uses an accurate vehicle model considering dead time and actuator limits.