Loose fruit collection using a delta robot

Abstract

With the increase of the global demand for palm oil, mechanizing oil palm plantations is becoming more and more significant. Even after almost a century since the first oil palm plantation in Malaysia, these plantations still rely almost-completely on manual labour for many of the harvesting and collection of the fresh fruit bunch (FFB) and the loose fruit (LF). Loose fruits that fall away from the FFB have the highest oil content since they are originally from the outer layer of the bunch. Due to its economic value, collectors have to manually collect these fruits which can result in many musculoskeletal disorders due to the bad back, hip and knee postures they have to endure on daily basis. Increasing the productivity and minimizing these issues of LF collection can be achieved by developing machines that can carry out the collection process seamlessly. Although many mechanisms like vacuum type collecting as well as racking tools have been implemented, each of which comes with constraints and issues like debris filtering and intensive manual labour. With the evolvement of robotics as well as machine vision and their applications in recent years, robotic collection of these fruits became more and more viable. In this work, automatic collection of these fruits using a delta robot along with a vision system for detection and localization is explored. A proof of concept of the collection process using a parallel delta link robot under a controlled environment is simulated. Additionally, a more versatile YOLO based LF detection system is explored. Although the robot displayed a capability of collecting over 50 LFs per minute, the integrated vision system for the delta robot proved incompetent when presented with different variations of LF and hence, a separate YOLO-based LF detection system is introduced.