Negative Obstacle Detection System
NREC developed a perception system to accurately detect negative obstacles in the path of an unmanned vehicle (UGV). Successful autonomous navigation requires detection not only of stationary and moving positive obstacles (guard rails, vehicles, people), but also the detection of negative obstacles (holes and ditches) at ranges that are nearly impossible for ladar and difficult for cameras.
NREC’s Negative Obstacle Detection System can detect negative obstacles with more certainty and at a greater distance so that UGVs can travel on safer and faster.
Detecting holes, ditches and other negative obstacles is more challenging than detecting walls, boulders, trees, moving objects and other positive obstacles. An accurate and dynamic negative obstacle detection system has several on- and off-road applications:
- Defense: greater UGV autonomy and higher speed will help warfighters quickly identify and avoid high-risk terrain and obstacles.
- Driver’s aid: enhancing human perception and alerting drivers to hazards will reduce the dangers of human error and fatigue.
- Agricultural equipment: ditches and holes out of the line of sight are particularly acute challenges to autonomous farming equipment.
- Emergency response vehicles: increased speed and safety are needed for more effective hazardous site access.
The prototype sensor system can detect a variety of negative hazards in day, night, and twilight operations. A 2D occupancy grid details the presence, location and type of negative hazards. The system is optimized for maximum vehicle speed and minimum Size, Weight and Power (SWaP). The modular system follows open standards and provide an open API to sensor data.
Far Range On-Line Learning
NREC employs a state of the art self-supervised machine learning technique called Far Range On-Line Learning (FROLL). It uses sensor data to train itself to identify objects observed at long ranges. This technique has already proven itself in off-road terrain perception and motion planning in the UPI program.