Optimising Fruit Harvesting Paths: A Mapless, Occlusion-Aware Picking Framework

Fruit harvesting is labor-intensive and increasingly challenged by the shortage of agricultural labor. To address viewpoint planning under occlusion, this paper proposes a mapless picking guidance framework that directly predicts the next viewing direction and estimates fruit occlusion without relying on pre-built maps or candidate-viewpoint sampling. Unlike conventional active vision methods that enumerate and evaluate multiple candidate viewpoints, the proposed method generates feasible viewpoints by jointly leveraging occlusion estimation and global picking direction supervision, thereby reducing computational cost and alleviating local optimum bias. An adaptive approach strategy is further introduced to balance viewpoint exploration and target approach during planning. Simulation results show that the proposed method achieves average success rates of 80.46% on the in-distribution test set and 77.58% on the unseen-fruit test set, with corresponding occlusion reductions of 80.56% and 79.16%, respectively. These results demonstrate the effectiveness of the proposed framework for occlusion-aware fruit viewpoint planning in unstructured orchard environments.