Inferring Implicit Trait Preferences from Demonstrations of Task Allocation in Heterogeneous Teams

Task allocation in heterogeneous teams often requires reasoning about multi-dimensional agent traits (i.e., capabilities) and the demands placed on them. However, existing methods tend to ignore the fact that not all traits equally contribute to a task. We propose an algorithm to infer implicit task-specific trait preferences in expert demonstrations. We leverage the insight that the consistency with which an expert allocates a trait to a task across demonstrations reflects the trait's importance to that task. Further, inspired by findings in psychology, we leverage the fact that a trait's inherent diversity among the agents controls the extent to which consistency informs preference. Through detailed numerical simulations and the FIFA 20 soccer dataset, we demonstrate that we can infer implicit trait preferences, and accounting for them leads to more computationally efficient and effective task allocation.