Autonomy Artificial Intelligence Robotics (AAIR)


by Paul H. Haley; Susan M. Thornton; Robert R. Mitchell; William P. Zachar; Mike Hoffelder; Steven McLean


Operating safely in cluttered environments is critical to future autonomous robotic operations as exemplified by FCS Risk 213. In support of this requirement, the Robotics Collaborative Technology Alliance (RCTA) program, sponsored by the Army Research Lab (ARL), has supported research tasks and corresponding integration and test events from 2006 through 2009. Multiple sensor systems, including scanning LADARs and stereo camera pairs, have been used to detect, track, and predict the future motion of obstacles in the close proximity of unmanned ground vehicles. These sensors produce frames of data at rates ranging from 6 to 30 Hertz. Resulting algorithm outputs are correlated to the local world and detection results both above and below the thresholds of the individual algorithms are recorded in a common format. This paper describes two methods for fusing the detection data. The first is a simplistic approach which implements a majority voting scheme amongst the algorithm results. The second, more rigorous, approach uses a “Strength-of-Detection” (SoD) method that utilizes an association step incorporating an error covariance model for each sensor, and also allows for cases where only a subset of the sensors report a detection. Results show that fused detection performance is far better than any single output due to the uncorrelated nature of single-sensor false alarms. We present representative results for both individual sensors and fused outputs.