Mechanizing the manual harvesting of fresh market fruits constitutes one of the biggest challenges to the sustainability of the fruit industry. During manual harvesting of some fresh-market crops like strawberries and table grapes, pickers spend significant amounts of time walking to carry full trays to a collection station at the edge of the field. A step toward increasing harvest automation for such crops is to deploy harvest-aid robots that transport the empty and full trays, thus increasing harvest efficiency by reducing pickers’ non-productive walking times. Given the large sizes of commercial harvesting crews (e.g., strawberry harvesting in California involves crews of twenty to forty people) and the expected cost and complexity of deploying equally large numbers of robots, this dissertation explored an operational scenario in which a crew of pickers is served by a smaller team of robots. Thus, the robots are a shared resource with each robot serving multiple pickers.If the robots are not properly scheduled, then robot sharing among the workers may introduce non-productive waiting delays between the time when a tray becomes full and a robot arrives to collect it. Reactive scheduling (e.g., “start traveling to a picker when the tray becomes full”) is not efficient enough, because robots must traverse large distances to reach the pickers in the field, thus introducing long wait times. Predictive scheduling (e.g., “predict when and where a picker’s tray will become full and dispatch a robot to start traveling there earlier, at an appropriate time”) is better suited to this task, because it can reduce or eliminate pickers’ waiting for robot travel. However, uncertainty is always present in any prediction, and can be detrimental for predictive scheduling algorithms that assume perfect information. Therefore, the main goal of this dissertation was to develop a predictive scheduling algorithm for the robotic team that incorporates prediction uncertainty and investigates the efficiency improvements in simulations and field experiments.
In the first part of this dissertation, strawberry harvesting was modeled as a stochastic process and dynamic predictive scheduling was modeled under the assumption that, once a picker starts filling a tray (a stochastic event), the time and location when the tray becomes full - and a tray transport request is generated - are known exactly. The resulting scheduling is dynamic and deterministic, and we refer to it as ‘deterministic predictive scheduling’ to juxtapose it against stochastic predictive scheduling under uncertainty, which is addressed afterwards. Given perfect ‘predictions’, near-optimal dynamic scheduling was implemented to provide efficiency upper-bounds for stochastic predictive scheduling algorithms that incorporate uncertainty in the predicted requests. Robot-aided harvesting was simulated using manual-harvest data collected from a commercial picking crew. The simulation results showed that given a robot-picker ratio of 1:3 and robot travel speed of 1.5 m/s, the mean non-productive time was reduced by over 90% and the corresponding efficiency increased by more than 15% compared to all-manual harvesting.
In the second part, the uncertainty in the predictions of tray-transport requests was incorporated into scheduling. This uncertainty is a result of stochastic picker performance, geospatial crop yield variation, and other random effects. Robot predictive scheduling under stochastic tray-transport requests was modeled and solved by an online stochastic scheduling algorithm, using the multiple scenario approach (MSA). The algorithm was evaluated using the calibrated simulator, and the effects of the uncertainty on harvesting efficiency were explored. The results showed that when the robot-to-picker ratio was 1:3 and the robot speed was 1.5 m/s, the non-productive time was reduced by approximately 70%, and the corresponding harvesting efficiency improved by more than 8.5% relative to all-manual harvesting.
The last part of the dissertation presents the implementation and integration of the co-robotic harvest-aid system and its deployment during commercial strawberry harvesting. The evaluation experiments demonstrated that the proof-of-concept system was fully functional. The co-robots improved the mean harvesting efficiency by around 10% and reduced the mean non-productive time by 60%, when the robot-to-picker ratio was 1:3. The concepts developed in this dissertation can be applied to robotic harvest-aids for other manually harvested crops that involve a substantial human-powered produce transport, as well as to in-field harvesting logistics for highly mechanized field crops that involve coordination of harvesters and autonomous transport trucks.