Bio-Inspired Visual Navigation: From Landmarks via Bearing to Controls

The proposed Bio-Inspired Visual Navigation System will enable UGVs to operate semi-autonomously with minimal input from an operator even in degraded communication environments. Semi-autonomous landmark-based navigation in low-texture indoor environments is difficult due to the lack of distinctive micro-features needed by conventional algorithms such as SIFT and SURF. In Phase I, Intelligent Automation, Inc. (IAI) and the GRASP Laboratory at the University of Pennsylvania demonstrated the feasibility of a novel solution comprising three main innovations. First, a fixation-based segmentation method is used to extract landmarks which are closed contours in the image and can be detected even in texture-less environments. All salient landmarks are automatically detected and tracked robustly under viewpoint and scale change. Second, landmarks are used as visual guides for bearing-based control of the robot to its destination. Third, an operator control unit (OCU) provides enhanced situational awareness and intuitive interaction between the operator and UGV. In Phase II, we will develop an integrated prototype and demonstrate system capabilities in a wide range of environments including inside buildings and tunnels. The system will be field tested on QinetiQ North Americas Tactical Robot Controller (TRC) and a TALON or Dragon Runner UGV.