The harvesting of tobacco leaves in China mainly depends on manual harvesting and self-transport of the harvester. The low level of mechanization, high labor intensity and high cost of tobacco leaf harvesting seriously affect the efficiency of tobacco leaf harvesting. Therefore, for the small-area and unstructured harvesting scenarios of tobacco fields, a predictive scheduling model for human-machine collaborative tobacco harvesting and transshipment was proposed based on a discrete-time hybrid system, which modeled the harvester's operation behavior as a discrete event, and then predicted its future transshipment demand, so as to drive the self-developed tobacco leaf transshipment robot to carry out active and forward-looking path planning, replacing the traditional passive response mode. Non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ) was used to solve the model and generate the corresponding human-machine collaborative optimization scheduling scheme. The simulation experiment verified the performance of the NSGA-Ⅱ algorithm in multi-objective optimization scheduling and the effectiveness of the predictive scheduling model of harvesting and transshipment by analyzing the influence of the ratio of the harvester and the transshipment robot and the threshold FR_tr of the harvester request on the harvesting efficiency. The results showed that when the number ratio of the transfer robot to the harvester was 1:2 and the FR_tr was set to 0.7, the non-productive operation time of the human-machine collaborative harvesting and transfer was reduced by 88.6% compared with that of the manual harvesting and transfer, and the harvesting efficiency was 95.7%. The results of simulated tobacco field experiments showed that when the number ratio of transfer robot to harvester was 1:1, the non-productive operation time of tobacco leaf harvest and transfer based on predictive scheduling strategy was 48.58 s, and the harvest efficiency was 83.1%. Compared with manual harvesting and reactive scheduling strategy, the harvest efficiency of tobacco leaf harvest and transfer was increased by 16.8 and 8.5 percentage points. When the number ratio of transfer robot to harvester was 1:2, the non-productive operation time of tobacco harvest and transfer based on predictive scheduling strategy was 53.14 s, and the harvest efficiency was 82.5%. The harvest efficiency was 14.6 and 9.9 percentage points higher than that of manual harvesting and reactive scheduling. The results can provide a reasonable and effective scheduling scheme for human-machine collaborative operation of tobacco leaves.