Multi-Robot simultaneous localization and mapping
Context
Simultaneous localization and mapping (SLAM) is an important issue in mobile robotics. Mapping describes the problem of integrating sensor inputs (sonar, range finders,… ) into a coherent description (a map) of the environment. Localization is the problem of determining the robot's position withing the environment. These two problems often cannot be solved independently. Using observations to build a map requires information on location of the robot to determine where the observation was made. Similarly, determining a robot's position in the environment is difficult without a map of its surroundings. As a result several SLAM algorithms have been developed to solve both these problems at the same time.
Goal
The objective of this thesis is to develop techniques for cooperative SLAM using multiple robots. In cooperative SLAM multiple robots collaborate to build a single map. The main advantage to this approach is that the robots can divide the work and achieve faster and more robust results by using sensor readings of multiple observers. However, to do this several issues need to be dealt with:
- Integrating sensor readings: an efficient way must be found to integrate sensor readings of all robots into a single map. Preferably, we would also like to minimize the communication required to build the global map.
- Determining the relative positions of the robots: When the initial positions of the other robots are unknown, the robots must first determine their relative positions before they are able to integrate their observations.
- Division of labor: develop strategies for efficient exploration, such that the amount of duplicated observations is minimized.
- Moving objects: when multiple robots are exploring the same environment, they may influence each other's sensor readings. As such the robots need a strategy for dealing with moving objects which cause different sensor readings to be made at the same location.
- Heterogeneous robots: in the most general setting we may employ different types of robots using different sensors. This introduces the additional difficulty that we must integrate different kinds of sensor data.
As part of the thesis the developed techniques will also be implemented on a set of real robots. For this we will use the Khepera robots available at the CoMo lab. To control the robots we will make use of the Player/Stage Framework.
