In task allocation a group of agents perform search and discovery of tasks, then allocate themselves to complete those tasks. Tasks are assumed to have a strong signature by which they can be identified. This paper considers task allocation in environments where the definition of a task is weak and can change over time. Specifically, we define tasks as environmental anomalies and present a new optimisation-based task allocation algorithm using anomaly detection to generate a dynamic fitness function. We present experiments in a simulated environment to show that agents using this algorithm can generate a dynamic fitness function using anomaly detection. They can then converge on optima in this function using particle swarm optimisation. The demonstration is conducted in a workplace hazard identification simulation.
展开▼
