Detection, tracking, and avoidance of moving objects from a moving autonomous vehicle

Abstract

ARL is developing the autonomous capability to directly support the Army's future requirements to employ unmanned systems. The purpose of this paper is to document and benchmark the current ARL Collaborative Technology Alliance (CTA) capabilities in detecting, tracking and avoiding moving humans and vehicles from a moving unmanned vehicle. For this experiment ARL and General Dynamics Robotic Systems (GDRS) conducted an experiment involving an ARL eXperimental Unmanned Vehicle (XUV) operating in proximity to a number of stationary and moving human surrogates (mannequins) and moving vehicles. In addition there were other objects along the XUV route of the experiment such as barrels, fire hydrants, poles, cones, and other clutter. The experiment examined the performance of seven algorithms using a series of sensor modalities to detect stationary and moving objects. Three of the algorithms showed promise, detecting human surrogates and vehicles with probabilities ranging from 0.64 to 0.85, while limiting probability of misclassification to 0.14 to 0.37. Moving mannequins were detected with slightly higher probabilities than fixed mannequins. The distance from the ground truth at the time of detection suggests that at a speed of 20 kph with a minimum distance to detection of 19.38 m, the vehicle would have a minimum of 3.5 seconds to avoid a mannequin or vehicle if detected by one of these three algorithms. Among mannequins and vehicles and, mannequins were more frequently detected than vehicles.