Localization and Motion Coordination Among Multiple Collaborating Mobile Robots
MetadataShow full item record
This dissertation mainly addresses the problem of multi-robot motion coordination, including deployment algorithm and some other important issues in the deployment process, for example, localization and collision avoidance. The first part of this dissertation is mainly about calibration and motion tests on the pioneer 3 DX mobile robot platform we are going to use. A pioneer 3 DX mobile robot has two driving wheels and one passive castor, and is a typical nonholonomic system. Due to nonholonomic constraint, the mobile robot has no nonzero side speed. In order to nicely control the robot and obtain its accurate running state, a motion calibration is necessary before serious experimental study. Three calibration parameters are adjusted manually until a well calibrated accuracy of motion is obtained. After calibration, a series of motion tests are conducted and the robot's capability of following commands is verified. Linear, rotational tests and a mixture of both movements are conducted in order, and the results show that the robot follows commands satisfactorily. The tests also provide some guides for our further experiments. Though encoder reading from robot is very accurate once the robot is well calibrated, due to accumulative error in encoders, a global localization solution is desirable. In the second part, a localization algorithm based on single camera is proposed and followed by an error analysis. The algorithm is based on trilateration and pinhole camera model. Error analysis indicates that the algorithm has very high accuracy of localization. In the next part, we discuss multi-robot deployment, which is a major of our work. A second order control method is proposed with our control frame work. This control framework belongs to the category of potential field methods. Compared with traditional deployment method that are based on position or velocity control, second order control is advantageous in making robots' movements smooth and natural. Besides, it is naturally incorporated in our control frame, which is originated from Hamilton's principle. By carefully designing the definition of artificial potential energy, the robots will move and be deployed automatically. Another issue in robot deployment is collisions during the process. In order to address this issue, collision avoidance schemes are proposed. The first collision avoidance scheme firstly determines relative movement between two robots. If the distance is below the setting dangerous distance and moreover they are still moving closer, the collision avoidance scheme will be triggered. The method used in our collision avoidance is to increase resistant force acting on robots to slow them down immediately. The second collision avoidance scheme calculate each robot's position in the whole team based on local neighbors' information, and then spread out the robots layer by layer-outer robots move firstly and inner robot move once enough space is left out. This scheme aims to reduce collisions, eliminate unnecessary movements and save power. The simulation results and experiments on Pioneer 3 DX mobile robots show that the deployment algorithm works well and the collision avoidance scheme can effectively reduce collisions during the deployment process. The last part of our work is a collaboration work we have done with IBM, and the objective of the work is to provide moving power for their current Mobile Measurement Technology (MMT), which is used to scan and get thermal map of data centers for power management purpose. By using a PatrolBot mobile robot, a Bumblebee camera, a wireless router and a laptop, the upgraded system did a successfully scanning of IBM's Southbury data center with an operator controlling MMT remotely. The new system saves labor and time, and is more convenient to operate compared with the old MMT. In the future, more work is on the way for the above several topics. Firstly, a sensor fusion method based on encoder and vision is desirable to obtain more accurate localization performance. Secondly, a more advanced and detailed study of collision avoidance is required to get a comprehensive understanding of collision avoidance. Thirdly, for MMT, a better user interface and more reliable localization method is also to be developed.