A3-[Multibot] - Research Phase 1 & 2

Phase 1

Although navigation for single robot systems has been studied intensively in the past, the problem of coordinating teams of mobile robots is virtually unexplored. One of the reasons is that the complexity of state estimation and planning problems generally increases exponentially in the number of robots, because one always has to consider the composite state space. On the other hand, the use of multiple robots is often suggested to gain advantages over single robot systems. Cooperating robots have the potential to accomplish single tasks faster than single robots. Additionally, teams of robots can be expected to be more fault-tolerant than a single possibly expensive robot. In this project we will investigate the problem of coordinating the actions of a team of mobile robots while they are exploring their environment. The key problem in multi-robot exploration is to choose different actions for the individual robots so that they simultaneously explore different areas of their environment. Since exploration is a dynamic process in which the robots acquire knowledge about their environment, the overall coordination task is a prototypical problem of spatial cognition. Throughout this project we will investigate different aspects of coordinating a team of mobile robots that explores its environment. In the first phase, we will consider appropriate spatial representations of the environment, improved targetpoint selection techniques, and decision-theoretic techniques for action selection under different situations of uncertainty. The long-term goal of this project is the development of an integrated technique for multi-robot navigation including multirobot exploration, multi-robot map registration as well as the maintenance of models by multi-robot systems acting in a dynamic environment.

Phase 2 

In the past, the problem of coordinating teams of mobile robots has been investigated intensively. Most of the approaches, however, are based on the assumption that the teams are homogeneous or that the whole team operates on a flat surface. First, the robots need an appropriate representation that allows them to store and maintain the relevant information necessary for navigating in their environment. Second, in the case of heterogeneous teams, the individual capabilities of the vehicles must be considered. For example, some of the vehicles might be legged robots which have a much richer repertoire of possible locomotion actions than typical wheeled systems and are able to traverse uneven terrain. At the same time, these robots might have different perceptual capabilities and different actuators. As a result, they have different information sources for modeling their environment, e.g., a legged system might be able to distinguish the substrates on which it is moving on the basis of its proprioceptive data. Computational limitations might prevent some of the systems from solving computationally intensive tasks such as mapping and localization. In addition, heterogeneous systems are likely to differ in regard to speed, operational range and energy consumption. This raises the question, how the individual capabilities can be considered when controlling the whole team of robots. Typical actions include perception actions for tracking other vehicles or actions for maintaining communication for data transfer have to be considered. Throughout this project, we will investigate probabilistic representations enabling a team of heterogeneous mobile robots to operate in terrains with potentially uneven and heterogeneous surfaces. Additionally, we will consider strategies for exploring such environments with robots that have different navigation capabilities. Throughout this project, we will develop a technique for representing non-flat environments that allow mobile robots to store and maintain different properties about the corresponding areas including surfaces and their traversability. We will also develop techniques for registration and localization based on this representation, which then can be used for building consistent maps of the environment. The additional information stored in the map will be utilized to improve the data association accuracy. Additionally, we will develop techniques for tracking robots from other platforms, so that more accurate pose estimates can be obtained. Finally, we will extend our decision-theoretic approach to mobile robot exploration developed in the first phase of this project so that it can deal with heterogeneous teams of robots using richer representations of the environment.