ASABE Chen 18 Detroit

Abstract

A small team of harvest-aiding mobile robots (FRAIL-bots) is being designed to aid large teams of human pickers in commercial strawberry harvesting by providing them with empty containers and transporting containers filled with harvested crops to collection stations at the edge of a field. The collective operation of these robots can serve requests for point-to-point transport in real time. The goal is to improve pickers’ job cycle by dramatically reducing non-productive walking and unloading times. To do so, individual FRAIL-bots with limited amounts must be dispatched and routed dynamically, in order to optimally match the dynamic and stochastic spatiotemporal distribution of real-time transport requests and resolve any spatial and resource sharing conflicts in the field correspondingly.

This problem corresponds to a partially stochastic and dynamic pick-up and delivery (DPD) problem. Dispatching of FRAIL-bots affects the overall job cycle because a picker is blocked from harvesting until (s)he receives an empty container and unloads the complete one to a robot. A real-time estimating model of picking pattern with uncertainty is monitored and estimated by a mechanistic model. The spatiotemporal distribution of future requests can be predicted based on picking pattern and observation of dynamic states. The Multiple Scenario Approach (MSA) has been shown to incorporate effectively stochastic and dynamic information in multiple vehicle routing problems. By scenario sampling, individual plans of different deterministic scenarios can be constructed and used to generate an optimized plan for a stochastic observation. The objective of an optimal plan is to minimize a) the total waiting time of all pickers to ensure the efficiency of field work, and b) the maximum waiting time of individual workers to incorporate fairness and increase acceptance of our system.

The MSA dispatcher will be tested using a simulator of manual strawberry harvesting. A basic heuristic algorithm that takes into account current requests only will be evaluated first. Then, the improvements achieved by the MSA method, which incorporates predicted future requests, will be assessed. For each individual deterministic sampled scenario, the exact solution is hard to obtain with exhaustive search, for real-time operation. Therefore, near-optimal solutions will be obtained from meta-heuristic algorithms.