||Novel Algorithms for Multi-Robot Path Planning, Task Coordination, Mapping and Localization in Indoor Environment
Indoor mobile service robots face many different problems to perform tasks au-tonomously. These problems become more complex when multiple robots are deployed in the environment for different services. This thesis addresses four important problemsfaced by multiple autonomous mobile robots in indoor environments.First, is the problem of smooth path generation for multiple mobile robots. Careful path planning is necessary for autonomous robots to navigate from their current position to the service location. A smooth and continuous path is desired for robot motion which avoids abrupt and sharp turns. This problem is further complicated in case of multiple robots as two or more robots may have intersecting and common paths which increases the chances of a robot colliding with another robot during its operation. Mul-tiple robots in the environment also poses problems of deadlock which must be resolved.The second problem which frequently occurs in multi-robots scenarios is the problem of common resource (like charging point or narrow path) sharing. Service robots for cleaning and patrolling are mostly in continuous operation and require frequent recharging. Using a large number of docking points proportional to the number of robots is costly and requires large space. A manager which can resolve conflicts and allocate the resource to the most appropriate robot by considering several factors like task priority, battery power left, distance travelled, etc. is required.Third, is the problem of autonomous task coordination and exploration by multiple mobile service robots. The problem is related to but more abstract than the first problem of smooth path generation, in which, the goal locations are given to each mobile robot. However, introducing multiple robots also introduce the problem of programming the robots to efficiently serve the region. The areas to serve in a map may vary with time. Moreover, the number of the robots available to serve may also be dynamic, in real world situations, as some of the robots may be charging, while some may be out of order. In case of robots used for surveillance, a robot may want other robot or robots to follow itself while chasing a suspicious person, for backup. This situation is also dynamic in terms of availability of robots, selecting the nearest robot for quick response, and selecting the same path towards a particular area as taken by the previous robot. In both the cases, explicitly programming the robots is cumbersome, and demands for a simpler scheme in which multiple robots can intelligently coordinate the tasks and explore the map. Moreover, if multiple robots are serving the same area, or navigating to the same area, they must coordinate tasks to maximize efficiency.The fourth problem is related to the third problem and relates to virtual pheromone deposition methodologies for multi-robot task coordination. Most of these approaches assumes pheromone deposition at perfect locations in the map. In reality however, it is difficult to achieve perfect localization of the robot due to errors in encoders and sensors attached to the robot, and the dynamics of the environment in which the robot operates. In real world scenarios, there is always some uncertainty associated in estimating the pose (i.e. position and orientation) of the mobile service robot. Failing to model this uncertainty would result in service robots depositing pheromones at wrong places. A leading robot in the multi-robot system might completely fail to localize itself in the environment and be lost. Other robots trailing its pheromones will end up being in entirely wrong areas of the map. This results in a ‘blind leading the blind’ scenario which reduces the efficiency of the multi-robot system.The final problem is a fundamental problem of Simultaneous Localization and Mapping (SLAM) in autonomous mobile robots. Feature based SLAM algorithm are popular which work by matching features like lines extracted from the sensors like camera or laser range sensors (LRS). Particularly, line detection is an important problem in computer vision, graphics, and autonomous robot navigation. However, in order to achieve efficiency in path planning, path smoothing, task distribution and other tasks, a robust SLAM algorithm is indispensable.
Hokkaido University（北海道大学）. 博士(工学)