Determination of apple attitude is a key technology for apple picking robots to achieve automatic picking. This paper proposes a joint estimation method for apple pose estimation by segmenting apples and their calyx basin and designs an improved YOLO11s segmentation network (SCH-YOLO11s) to address challenges posed by small, darker calyx basin targets and image degradation. The SCH-YOLO11s network combines the Simple Attention Module (SimAM) with C3k2 into the C3k2_SimAM module, the Conv in the backbone network is replaced with the CMUNeXt Block, and the Histogram Transformer Block (HTB) is added to the C2PSA module. The trained model segmented the apple and the calyx basin and acquired the point cloud data of the segmented region. The center of the apple point cloud was determined by least squares sphere fitting, and the center of the calyx basin point cloud was calculated using the mean value method. The vector connecting these two centers was defined as the apple’s pose. The SCH-YOLO11s network achieves a segmentation AP50 of 97.1% and 94.7% on the apple and calyx basin, and the mAP is improved by 1.8% and 2.7% compared to the unimproved version, respectively. Real apple pose data were obtained for experimental comparison with the estimated pose data. The average error angle of the real pose data compared with the estimated data is 12.3 degrees. The algorithm’s runtime per image is approximately 0.08 s. It shows that the proposed pose estimation scheme has the capability to be applied in a real apple picking robot system.

In the field of orchard harvesting, this study proposes a collaborative picking strategy for dual-arm robots, aiming to improve efficiency, reduce labor burden, and achieve precise automation. The strategy combines the Multi-Agent Proximal Policy Optimization (MAPPO) algorithm with the Multi-Objective Greedy Picking Strategy (MOGPS) algorithm. By centrally training the critic network and decentralizing the actor network, the robots can autonomously learn and precisely pick in a simulated environment. To address dynamic obstacle avoidance, a dynamic collision assessment strategy is proposed, and an improved MOGPS algorithm is used to consider the distribution of fruits and the complexity of the working environment, achieving adaptive path planning. Experimental results show that the MAPPO-MOGPS algorithm optimizes the picking path by 15.11%, with a picking success rate as high as 92.3% and an average picking error of only 0.014. Additionally, physical experiments in real-world settings demonstrate the algorithm’s practical effectiveness and generalization.