Task Allocation for Heterogeneous Robots Using a Self-Organizing Contextual Map

Abstract

The Self-Organizing Contextual Map (SOCM) is applied to address the task allocation problem in the context of multi-robot heterogeneous systems. Inter-variability and overlap of common capabilities (sensors, actuators, and descriptors) exist between robotic agents. As a starting point, a range of search-and-rescue tasks are used to form a set of binary symbol codes that identify selected tasks. Binary vectors are then predefined to represent factitious features (task requirements) to solve these search-and-rescue tasks, forming the attribute codes. The combination of these vectors is used to form associations between the features of a task (requirements) and a robotic agent's capabilities. Using a recursive stepwise approximation and calculating the Euclidean distance, feature and contextual maps are formed to differentiate between tasks. Then similar predefined binary vectors are used that represent the robotic agent's capabilities. Using this information, nested domains are created on top of the contextual map that form allocation regions for all learned tasks. Three generalization tests are then performed to assess the SOCM's robustness and task allocation abilities. Overall, the results suggest that the applied SOCM encapsulates a true heterogeneous system for addressing a simplified version of the task allocation problem.