Multi-armed Bandit Approach for Task Scheduling of a Fixed-Base Robot in the Warehouse

Ajay Kumar Sandula, Pradipta Biswas, Arushi Khokhar, Debasish Ghose

Abstract: Robot task scheduling is an increasingly important problem in a multi-robot system. The problem turns more complicated when the robots are heterogeneous with complementary capabilities and must work in coordination to accomplish a task. This chapter describes a scenario where a fixed-base and a mobile robot with complementary capabilities accomplish the ‘task’ of moving a package from a pickup point to a shelf in a warehouse environment. We propose a two-fold optimised task scheduling approach. The proposed approach reduces the task completion time based on spatial and temporal constraints of the environment. The approach ensures that the fixed-base robot reaches the mobile robot exactly when it brings the package to the reachable workspace of the robotic arm. This helps us to reduce the waiting time of the mobile robot. The multi-armed bandit (MAB) based stochastic task scheduler considers the history of the tasks to estimate the probabilities of corresponding pickup requests (or tasks). The stochastic MAB scheduler ensures that the mobile robot with higher estimates of probabilities is given top priority. Results demonstrate that a stochastic multi-armed bandit based approach reduces the time taken to complete a set of tasks compared to a deterministic first-come-first-serve approach to solving the scheduling problem.