An implementation of a novel localization framework for robots and its application to multi-robot tasks

Abstract

In recent years, the trend in robotics is to replace single robot platforms with teams of robots that cooperate to achieve a common goal. The main reasons are the fact that single robots may break down leading to a complete failure of achieving a goal and that certain tasks, such as chemical plume tracing, distributed sensing grids, or map generation, are difficult, if not impossible, to be solved with only one robot. Replacing a single robot by a team of robots can lead to a system that is fully distributed, more robust, noise tolerant, and highly scalable. One of the bottlenecks in the development of successful collaborative robotics is an enabling localization technology that shares the design criteria of collaborative robots, as we mentioned above. Existing localization technologies rely either on global information provided by GPS, beacons, and landmarks or on complex local information provided by vision systems. Using global information limits the use of the localization to specific environments and it is more prone to errors. Vision-based localization systems require expensive hardware and their accuracy is limited both by distance and lighting conditions. To address the problems with previous technologies we propose a novel plug-in 2D localization hardware module for multi-robot localization, based on trilateration, which is fully distributed, inexpensive and scalable. In 2D trilateration, to locate a remote robot, the sensing robot must know the locations of three “base points” in its own coordinate system and be able to measure distances from these three points to the remote robot. Our distance measurement method exploits the fact that sound travels significantly slower than light, allowing us to employ a Difference in Time of Arrival technique. Let us assume that we have two robots that are both equipped with our localization module. Each localization module has one radio frequency (RF) transceiver and three ultrasonic acoustic transceivers, which are also the “base points”. Suppose robot 2 (remote robot) simultaneously emits an RF pulse and an acoustic pulse. When robot 1 (sensing robot) receives the RF pulse (almost instantaneously), a clock on robot 1 starts. When the acoustic pulse is received by each of the three acoustic transceivers on robot 1, the elapsed times are computed and converted to distances. Robot 1 is now able to use trilateration to compute the location of robot 2. Similarly, robot 2 can localize robot 1 when robot 1 emits the RF and acoustic pulse. To test our localization hardware we have developed a robot platform which we called Maxelbot. The modular design of this platform allows for easy addition of hardware modules, as we present later in this thesis. Next, we demonstrate the performance of this technology with various empirical tests on moving and non-moving platforms, both in controlled indoor environments and outdoors. We further test our localization module by using it to monitor the performance of a novel algorithm for uniform coverage of a region. Surveillance and mine detection are some applications that would benefit from this algorithm. Prior work has claimed that this task is impossible to solve for non-convex regions (Gage 1993). Our algorithm enables robots to uniformly cover regions of all shapes, such that the robot movements are not predictable and the region periphery is not neglected. The algorithm assumes that robots are independent and is physics-based, relying on an analogy with mean free paths of particles. Validation of the algorithm is rigorously provided via simulation and Maxelbot experiments. Finally, we demonstrate how our localization technology allows us to tackle an important problem in cooperative robotics – the self-organizing of chain formations in unknown environments. This algorithm assumes that robots are dependent and is also physics-based, using a surprisingly simple andelegant modification to the standard Artificial Physics framework (Spears