THE NEW ADVANCED PROTOTYPE OF AIRBORNE VISUAL CONTROL OF A GROUND ROBOT
Keywords:Vision-based navigation of ground robots, Airborne control from tethered platform, Deep learning convolution network, Time delay of autopilot, Stability of differential equations
Unmanned autonomous robots will be widely used very soon for land use, treatment, and monitoring. The paper addresses the problem of visual navigation of ground robots using a camera positioned at a certain elevation above the confined area. The main idea is that robot’s “eyes” is not located on the robot but are independent autonomous system. As a result, the “eyes” can go up and observe the robot from above. This patented technology of airborne vision-based control of ground robots is already described previously in "International Journal of GEOMATE" and Geomate2019 conference paper (Ehrenfeld et al., 2020). There was described in detail the small toy prototype with very simple algorithms and very shortly the next more complex prototype. In the current paper detailed algorithms for navigation and control of this complex prototype is presented. The system consists of the USB camera on the tower (4 meters) connected to the computer which controls the big ground robot by sending Wi-Fi commands. For navigation (recognition of the ground robot, finding its position and orientation) was used deep learning (artificial neural network). It was demonstrated that this robot can pass two main tests: travelling along some predefined path and random walk inside of some predefined boundary.