Autonomous Non-communicative and Vision-based Control Strategy for AGV to Track UAV

UAV-AGV (Unmanned Aerial Vehicle-Autonomous Ground Vehicle) is a suitable system for automation of industrial tasks such as inspection and mapping. Vision-based UAV-AGV system (camera attached to UAV or AGV or both) is prominent in GPS inaccessible zones. Vision-based tracking motion between both agents is an important aspect during execution of outdoor autonomous missions. It can be either UAV tracking AGV or vice-versa. Former one (UAV tracking the motion of AGV) is well solved. Latter one is less focused as per the literature. But it also required for effective collaboration of the two-agents. For UAV leader and AGV follower roles, AGV needs to continuously track UAV for supporting (payload and tethered charging) longer outdoor missions of UAV. However, existing work in the literature is not suitable for the outdoor missions. Hence, this work presents a non-communicative and vision-based tracking strategy for AGV to track the motion of UAV. AGV contains an on-board sky-facing camera to localize UAV. Deep learning-based technique is implemented for robust localization of UAV using on-board camera of AGV. Sliding mode control technique is implemented to develop tracking controller for AGV. UAV is manually operated with random velocity commands and AGV autonomously tracks the projection of center of UAV on the ground plane. Tracking experiments are performed using quadcopter UAV and non-holonomic AGV mobile platforms. Learning based detection of UAV and tracking controller of AGV are showing satisfactory performance from the experimental results. This shows feasibility to execute outdoor missions without requirement of GPS and also communication between the agents.