A Compact Visual-Odometry Sensor for UGVs

Small, Unmanned Ground Vehicles (UGV) will play a significant role in future army applications including reconnaissance, mine detection, explosive ordnance disposal and resupply. Typically, a UGV operates in concert with other agents, including: UGVs, humans, and/or manned vehicles. To successfully navigate in the field, UGVs require reliable odometry. Mechanical odometry is fraught with wheel slip and skid issues and inertial solutions tend to drift over time. Visual odometry uses imagery of the surrounding terrain, acquired over time, to deduce vehicle motion. Salient features are acquired and tracked, from frame to frame, as the vehicle passes them. Natural terrain is particularly suited to this technique as features tend to be well distributed. We propose an innovative visual odometry sensor for use over rough terrain, which requires far less power and is more compact than anything produced to date. The sensor will embody the integration of TRACLabs’ embedded vision processing technologies with innovative visual odometry algorithms developed by Dr. Frank Dellaert of Georgia Tech.. A key feature of the system will be its turn-key operation. Field calibration will not be required to yield continuous 6DOF (Degree-of-Freedom) output. The system will be self-contained and not require coordination with other on-board sensors.