STB-PHD: A Trajectory Prediction Method for Symmetric Center-of-Gravity Deviation in Grasping Flexible Meat Cuts

In automated sorting and grasping of livestock meat cuts, the ideal assumption of symmetric mass distribution is often violated due to irregular morphology and soft tissue deformation. Under the combined effects of gripping forces and gravity, the originally balanced configuration evolves into an asymmetric state, resulting in dynamic shifts of the center of gravity (CoG) that undermine the stability and accuracy of robotic grasping. To address this challenge, this study proposes a CoG trajectory prediction method tailored for meat-cut grasping tasks. First, a dynamic model is established to characterize CoG displacement during grasping, quantitatively linking gripping force to CoG shift. Then, the prediction task is reformulated as a nonlinear state estimation problem, and a Small-Target Bayesian–Probability Hypothesis Density (STB-PHD) algorithm is developed. By incorporating historical error feedback and adaptive covariance adjustment, the proposed method compensates for asymmetric perturbations in real time. Extensive experiments validated the effectiveness of the proposed method: the Optimal Sub-Pattern Allocation (OSPA) metric reached 4.82%, reducing the error by 4.35 percentage points compared to the best baseline MGSTM (9.17%). The task completion time (TC Time) was 6.15 s, demonstrating superior performance in grasping duration. Furthermore, the Average Track Center Distance (ATCD) reached 8.33%, outperforming the TPMBM algorithm (8.86%). These results demonstrate that the proposed method can accurately capture CoG trajectories under deformation, providing reliable control references for robotic grasping systems. The findings confirm that this approach enhances both stability and precision in automated grasping of deformable objects, offering valuable technological support for advancing intelligence in meat processing industries.