A Modular Framework for Multi-robot Localization Scenarios

Localization, in mobile robotics, is the ability of the vehicles to discern their position and orientation into their environment. Today, there are two main approaches to the localization problem. Deterministic techniques use recursive estimation algorithms to determine the robots state vector directly while at the same time filtering out potential noise, which corrupts the system. Alternatively,. probabilistic techniques recursively update the probability of a mobile robot to occupy a particular position in space. Both of these contrasting techniques exhibit particular advantages and disadvantages in real-time implementation. For example the deterministic algorithms have less memory and processing requirements from their counterparts, and that makes them ideal for real-time applications. This thesis presents a modular localization framework based on extended Kalman filter estimators, which is applied in both single and multi robot scenarios. The modularity of the framework is demonstrated in the multi-robot scenarios where robots can be easily added and removed from the group in real-time with only changing (degrading or upgrading) the system performance but not disabling it completely. The key idea for performing collective localization is that the group of robots must be viewed as a single centralized system where a,Centralized Extended Kalman Filter (CEKF) is implemented to track the sequential positions of a number of robots in an initially known environment. The system is tested for situations ~ where the robots have access to al1solute environmental landmarks and where the robots rely only on relative distance and bearing measurements. Furthermore, methods to parallelize the CEKF algorithm and distribute it amongst the number of robots that occupy the team are demonstrated and teste'd. This technique can be used for real-time implementation scenarios where the designer needs to take full advantage of the robots' local processing capabilities, which otherwise would be impossible with the centralized form ofthe algorithm.EThOS - Electronic Theses Online ServiceGBUnited Kingdo