Towards Autonomous Apple Fruitlet Sizing with Next Best View Planning

Harry Freeman, George Kantor

Keywords: Agricultural Robotics, Fruit Sizing, Next Best View, Viewpoint Planning

TL;DR: A next best view planning approach for sizing apple fruitlets in agriculture with preliminary results in simulation.

Abstract: As a result of the expected effects of global population growth and the anticipated need to increase food production, agricultural robotics has become a popular research area. Robots are able to automate laborious and time-consuming tasks which allows farmers to make faster decisions and ultimately improves yield. In this paper, we build towards designing a robotic system that autonomously sizes apple fruitlets. Our proposed system adopts a viewpoint planning approach targeted towards sizing smaller fruit. We utilize a coarse and fine planning tree along with a region of interest utility-gain mechanism to generate next-best view candidates to capture images of fruitlets. A truncated signed distance function is used to build a dense surface point cloud and fruits are sized using a combination of 3D and 2D techniques. We provide preliminary simulated results demonstrating that our system can effectively size fruitlets in occluded environments.