Predicting Tomato Bioyield Force Using Compression Stiffness and Complementary Fruit Properties for Robotic Grip Force Planning

Fruit-specific estimates of failure-related force could support grip force planning in tomato harvesting robots. This study evaluated whether the apparent compression stiffness obtained during pre-bioyield low-load compression can predict the individual mechanical bioyield force. The prediction was assessed through post-test evaluation using the complete compression record and through sequential replay with prediction-based stopping. The contribution of complementary fruit properties, including fruit weight, equivalent diameter, and Jeffreys model viscoelastic parameters, was also evaluated relative to the use of stiffness alone. The main analysis included 87 of the 103 tested “Tarffy 17” tomatoes. Stiffness was calculated across 30 absolute-force intervals and evaluated using leave-one-out cross-validation. In the post-test, the lowest mean absolute error (MAE), 9.6 N, was obtained for [0,30) N. In the sequential replay, the lowest MAE, 10.2 N, was obtained for [5,35) N, whereas using the [0,30) N interval increased the MAE to 35.9 N. In the post-test LOOCV, adding fruit weight, equivalent diameter, or Jeffreys model viscoelastic parameters resulted in only minor changes in MAE, with no significant differences in paired errors. These results show that low-load stiffness alone provided fruit-specific information for predicting mechanical bioyield force under the present conditions and that the stiffness calculation interval and prediction timing should be considered jointly for sequential use.