Download or read book A Human-Animal-Robot Cooperative System for Anti-Personal Mine Detection written by Thrishantha Nanayakkara and published by . This book was released on 2008 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: To the best of the author's knowledge, this is the first time a human-robot-animal integrated system is tested for antipersonnel landmine detection. The proposed system tries to integrate distinct capabilities of three different systems to improve the effectiveness of landmine detection in a cluttered environment. The mongoose is found to be a rodent with extremely sensitive olfactory capabilities, dexterous navigation capabilities in a cluttered environment, and small enough to burrow through rubble. The lightweight legged robot (4kg) can move in a minefield without detonating landmines, carry a metal detector, and interact with the mongoose and the human. The remote human operator can analyze the behaviors of the animal-robot system and judge how best the system should move from a remote location. Therefore, the system achieves a fundamental objective of humanitarian landmine detection by improving the effectiveness and accelerating the detection process through removing the human operator from the minefield. The design gave much emphasis on reducing the need to have expensive sensors and sophisticated image processing systems in order to make it as cost effective and reliable as possible. Therefore, there were only a single sonar proximity sensor and two bumper switches attached to the front of the robot. However, further improvements are needed in the arbitration mechanism that optimizes the synergy among the human, robot, and the animal by improving the learning algorithms. The robot can learn from both the animal and the human though the teaching signals can be noisy. The animal can learn from both the human and the robot to navigate with the robot attached to it. The human can learn from the animal and the robot by observing the limitations of the animal-robot system. We are conducting further research on learning algorithms that suits this scenario. Commensurate efforts have to be taken to simplify the learning algorithms to suit commercially available embedded processors and to improve the processor network to accommodate the extra processing load. Furthermore we hope to automate the training process of mongooses based on the wealth of knowledge we have gathered through manual training. This will allow the trainers to run the training sessions round the clock.