Virtual Reality-Based Active Vision Teleoperation for Sweet Pepper Inspection in Occluded Environments

In agricultural environments, foliage occlusions present a significant challenge for crop inspection. While fully autonomous systems often struggle to generalize across these occluded conditions, human operators possess an inherent cognitive and visual adaptability that provides a valuable alternative for navigating such complexities. This work introduces an approach for crop inspection that combines Virtual Reality teleoperation with an Active Vision strategy to address occlusion limitations in agricultural environments. This specific configuration, leveraging human-driven active vision for detailed crop inspection, has not yet been widely explored. The system enables users to remotely explore a simulated crop plant through natural body and head movements, while a robot arm equipped with an RGB camera replicates these motions in real time. The proposed system was experimentally validated in a controlled laboratory setup using real sweet peppers with artificial leaves, where 30 participants without prior knowledge of the system performed peduncle localization tasks. Cycle completion time, Peduncle Detection Interaction (PDI) recall and precision, and subjective task workload (NASA-TLX) were evaluated. The results demonstrated an average cycle completion time of 12 s for each sweet pepper peduncle, a PDI recall of 96.7%, a PDI precision of 92.1%, and a mean NASA-TLX score of 32.7. Overall, the findings indicate that the proposed Virtual Reality-based Active Vision framework offers an intuitive and efficient means of viewpoint control, demonstrating its potential as a human–robot interaction solution for agricultural inspection tasks.