ABSTRACT
This book chapter highlights the control of mobile manipulators with object detection for explosive ordnance disposal (EOD) application. The proposed design of a remotely operated mobile manipulator can be useful in finding suspicious activity and environment. The robot prototype is an unmanned ground vehicle (UGV) with a multi-terrain tracked wheel chassis and a 6DOF robotic arm mounted on the top along with an FPV camera for visual feedback. The robot operation is performed wirelessly from a remote computer or a mobile device to assist bomb disposal teams in safely detecting the type of explosive device and disposing of it remotely with help of the manipulator. Deep learning methods are used for real-time object identification and detection using an open-source framework called TensorFlow object detection API, using the SSD Mobilenet V2 trained on the MS-COCO dataset. A custom dataset for detecting explosive devices and suspicious objects has been prepared by using photographs of various explosive devices, IEDs, bombs, and wrapped packages that are likely to create suspicion. The custom images are annotated and labeled into a pre-trained SSD model for custom object detection. The bomb or an explosive device that has been identified is reported to the operator for manual interaction using the robotic arm. The detected bomb then can be tactically diffused, contained, or disposed of at a remote location. The prototype is tested by performing remote teleoperation 118on a rough terrain also the custom object detection model developed using the SSD Mobilenet V2 model is tested for a random package detected as an unidentified package as test results.
