Presentation
Papers
Overview
Multi-robot systems is a challenging field that has been receiving increasing attention in the robotics community over the past years, mainly because a group of robots can potentially be more robust and efficient, present increased fault-tolerance, and even be able to accomplish tasks that cannot be executed by a single robot.
However, to achieve the full potential in these systems, robots must be able to cooperate with each other. As the number and complexity of robots increase, one of the problems that emerge is how to distribute tasks among the members of the group. This problem is also known in the literature as the Multi-Robot Task Allocation problem (MRTA).
In this work we develop a new framework to tackle the MRTA problem applied to a group of heterogeneous multi-task mobile robots. A multi-task robot is one capable of executing multiple tasks simultaneously in contrast to most of related approaches that assume single-task robots, i.e. robots capable of executing at most one task at a time, which restrict the capabilities of the robots. The main contribution of this work is the development of a framework that makes it possible to mobile robots execute multiple tasks simultaneously, taking into account constraints like pose, power consumption, and communication bandwidth. The approach is based on the assumption that a task can be separated into smaller pieces, which we call actions, that can be executed in different robots.
The approach can thus be divided into four steps: (i) define, for each task, a set of actions that can accomplish the task's goal; (ii) design all actions such that the inputs can be modeled as queries in a distributed database, which is robot and sensor independent; (iii) assign, for each robot, a set of constraints that will capture all restrictions necessary to determine if two or more actions can be executed concurrently; (iv) develop a task allocator which will obey the constraints and maximize some pre-defined criteria.
We model an action such that the input data can be decoupled from the robot or sensor that produces it. This means that if an action requires a given type of data, it is not necessary that it is acquired by the same robot where the action is running, but can be provided by any robot that meets its specifications. Furthermore, each action defines a set of constraints imposed over the robot's share-restricted resources like position, power consumption and communication bandwidth. This will allow us to evaluate if two actions can be executed simultaneously, making it possible to design multi-task robots, i.e. robots capable of executing multiple tasks concurrently.
We named the proposed MRTA framework as CoMutaR (Coalition-formation based on Multitasking Robots). A coalition is a temporary organization of actions that are brought together in order to execute a task which is dissolved once the task is accomplished. Each coalition is responsible to execute one task and, since the data source for an action can be easily switched among sources, this framework can provide a more flexible and fault-tolerant solution to the reallocation problem.