Harvest-order planning for a multiarm robotic harvester

Abstract

A multiarm robotic harvester is being developed for two-dimensional crops such as melons. A number of Cartesian manipulators, mounted in parallel on a rectangular frame, traverse laterally across the crop bed as the frame moves along it. The robotic arms reach down to pick melons and place them on adjacent lateral conveyors. The coordinates of the fruits to be harvested are assumed to be known prior to harvest so that the robot gets a bank of targets in local coordinates. In this paper, we describe the algorithms developed and used to plan the assignment of melons to be harvested by each of a number of arms in a collaborative way so that the maximum number of fruits will be harvested by a given number of arms. Under practical kinematic conditions, the fruits and the manipulators’ capabilities can be modeled as a task of coloring an interval graph, and a greedy algorithm known to produce an optimal solution for a k colorable sub-graph problem is used. Under faster manipulator performance, an approximation algorithm based on heuristics and a local search was shown to produce near-optimal harvest assignments. The algorithms are used to facilitate the design of the robot using simulations of the effects of robot speed, number of arms, manipulator’s lateral acceleration and fruit handling time on the harvest. The simulations enable economic optimization of the design of such robotic harvesters, taking into account the costs of robotic arms, labor and operation time and the value of the crop.