Abstract: Taro is one of the three major export specialty aquatic vegetables in China (along with lotus root and water chestnut). It can play an important role in the national strategy of rural revitalization. However, the current taro harvesting can rely mainly on manual labor, with the labor cost accounting for 40% to 50%. Mechanical harvesting has been confined to the conical spherical root-soil composite structure under a wet and sticky soil environment. The root system is also entangled and adheres to the soil during harvesting. It is then difficult to separate the taro from the soil. At the same time, it is prone to mechanical damage when subjected to impact, due mainly to the high moisture content (as high as 73.70%) of taro during harvesting. Therefore, it is urgent to explore the root-stem soil separation and the optimal separation mode of transportation in the current mechanical harvesting of taro. Particularly, the traditional rod-bar lifting chain can also be limited to the transport problems, such as the root-soil composite body rolling back and falling, low transportation efficiency, and high damage rate of taro. In this study, a flexible baffle-type device was designed for the taro transportation separation. The device was composed of rods, rubber baffles, vibrating wheels, frames, and chain drives. The collision dynamics were also analyzed on the disintegration and throwing of the flexible baffle on the root-soil composite body. The influencing factors on the transportation separation performance were identified as the height of the flexible baffle, the lifting line speed, the inclination angle of the screen surface, the vibration frequency, and the amplitude of the vibrating wheel. Discrete element method - finite element method - multi-body dynamics (DEM-FEM-MBD) were combined to determine the effects of the factors on the average transportation time and soil screening rate of taro. A series of experiments was finally carried out using a single-factor simulation. The results showed that the flexible baffle effectively prevented the root-soil composite body from rolling back. The maximum impact force of the flexible baffle on the taro was 55.66% lower than that of the rigid ones. The spacing of the flexible baffle was determined to be 300mm. The mesh size of the flexible baffle was determined to be 4mm using grid convergence. The relative error between the fitting curve and the physical experiment was 1.78%. A Box-Behnken response surface bench test was carried out to explore the influence of the transportation screen inclination angle, lifting line speed, and vibrating wheel frequency on the average transportation time and soil screening rate of taro. An optimal combination of the parameters was determined after multi-objective optimization: A transportation screen inclination angle of 18°, a lifting line speed of 0.62 m/s, and a vibrating wheel frequency of 2 Hz. The soil screening rate was 88.76%, and the taro transportation time was 1.91 s. The relative errors between the coupled simulation and the bench tests were 0.20% and 3.08%, respectively, which verified the reliability of the coupled model. Field verification showed that the soil screening rate and transportation success rate increased by 4.72 and 26.67 percentage points, respectively, in the flexible baffle-type transportation separation device. While the damage rate and harvest loss rate of taro were reduced by 12.11 and 0.97 percentage points, respectively. The separation and harvest quality of the flexible baffle-type transportation separation device were superior to those of the traditional rod-bar ones, fully meeting the requirements of the multi-sub-taro harvesting. The findings can also provide a strong reference to optimize the efficient and low-loss harvester for the taro, root, and tuber crops.