Demand-Aware Multi-Robot Task Scheduling with Mixed Reality Simulation

Подробнее о видео

This paper addresses the problem of multi-robot task scheduling by estimating the demand for the tasks in a real-world scenario. Scheduling tasks for multiple robots becomes complex when a human is involved in allocating limited resources. We propose a stochastic multi-agent multiarmed bandit based task scheduler which prioritizes the tasks
based on the estimated demand for the tasks. To gain insight into the varying priorities of a human task allocator in a multi-armed bandit scenario, we conducted a user study in a Mixed Reality environment which can be used to customize the resource allocation process. We observe that the users consistently made sub-optimal choices due to their preference
to minimize other parameters of the real world scenario rather than strictly adhering to the optimal strategy. Our proposed method uses the Thompson Sampling bandit algorithm with ϵ-greedy approach to solve the multi-agent multi-armed bandit problem. The approach outperformed other methods such as first-come-first-serve, rate monotonic scheduling, and heuristic based Min-interference approaches in terms of the “demand aware performance index