IMPLEMENTATION OF SMART NAVIGATION ON RESCUE ROBOT FOR POST EARTHQUAKE RUINS EXPLORATION

• Main Author: Zakie Shahab, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82439
• Institution: Institut Teknologi Bandung

Evacuation process beneath ruins of buildings after an earthquake poses some problems to the rescue team due to risks that may happen either during search from above or below those ruins. Addressing to this issue, a rescue robot has been developed to search victims beneath the building ruins. This robot comprises a power system as energy supply to the robot, a navigation and propulsion system as the main body of the robot, and an interface used to monitor the robot’s movements. This robot operates autonomously using spiral algorithm and reinforcement learning. Spiral algorithm in this smart navigation is used for partition exploration, meanwhile the reinforcement learning is used as retracking and backtracking method. Reinforcement learning is a machine learning method in which the agent does not have prior data during training and generates its data based on the actions taken during training. In order to perform an optimal smart navigation, the robot must have components capable of detecting the location of obstacles in its surroundings. By identifying the location of these obstacles, the smart navigation algorithm can avoid collisions. Additionally, this navigation algorithm ensures the robot to cover whole area accessible to its movement, considering that the exact location of victims beneath the rubble is unknown. Therefore, the robot needs to traverse the entire building to locate all victims and facilitate comprehensive rescue planning. Looking at the result of smart navigation performed by the robot using several dummy arenas, the robot's performance was evaluated in terms of coverage ratio and path efficiency. The algorithm achieved a coverage ratio of 1.09 and a path efficiency of 0.92, with an optimal value being 1. Retracking with the reinforcement learning algorithm is initiated when the spiral algorithm has no further areas to explore, and backtracking using reinforcement learning occurs when the entire reachable area has been explored, returning the robot to its initial position. Additionally, the robot's movement map and the positions of the victims can be viewed on the interface provided. Using the map and the distribution of victim positions, the rescue team can plan evacuations more effectively.