Roboharvest: AI-driven robotic system for automated loose palm oil fruitlet detection and collection in plantations

Malaysia and Indonesia remain leading global palm oil producers, with Malaysia's output projected at 19.7 million tonnes for 2023–2024. Despite a century of cultivation, the palm oil industry heavily relies on manual labor to harvest fresh fruit bunches (FFB) and collect loose fruits (LF). LF, the outer layer with the highest oil content, holds significant economic value. However, manual collection exposes workers to musculoskeletal disorders due to prolonged poor postures. This study explores automating LF collection using a cost-effective robotic system with an integrated vision system. A proof of concept employs a 6-degree-of-freedom (DOF) robotic arm controlled by a Raspberry Pi microprocessor within a simulated environment. The system utilises YOLOv10-based object detection and image segmentation, coupled with Particle Component Analysis (PCA), in a Robot Operating System (ROS) framework for precise LF localization. Predictive inverse kinematics using Recurrent Neural Networks (RNN) ensures dynamic adaptability during collection. Preliminary results show the robot collects approximately 2 LF per minute due to hardware limitation, demonstrating its potential to enhance productivity and reduce physical strain on workers. However, advanced processors can significantly improve its performance