Machine Learning and

Computer Vision Integration

Lidar System and Point Cloud Processing

Onboard lidar system identifies and localizes potential landing sites through RANSAC plane-fitting and point cloud clustering.

Stereo Camera Integration

Stereo camera system uses a custom trained YOLOv8 segmentation model to identify and localize approved landing surfaces (e.g. brick walls & house walls)

Data Association

Lidar bounding boxes of landing-sites and Stereo Camera bounding boxes landing-sites are compared using IoU to confirm and produce list of viable landing locations.

Deep Reinforcement Learning

Landing maneuver strategies are learned through Deep Reinforcement Learning in custom Gazebo simulation with ROS