Pomobot: a robot for autonomous tomato harvest

• Grant number: 23/13465-3
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: April 01, 2024
• End date: December 31, 2024
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Pedro Viganó da Silva Santos
• Grantee: Pedro Viganó da Silva Santos
• Company: BSV ROBOTICS LTDA
• CNAE: Desenvolvimento de programas de computador sob encomenda ; Serviços de engenharia
• Pesquisadores principais: Nicholas Smaal
• Associated researchers: Roseli de Deus Lopes
• Associated scholarship(s): 24/03998-7 - Pomobot: a robot for autonomous tomato harvest, BP.PIPE; 24/04009-7 - Pomobot: a robot for autonomous tomato harvest, BP.PIPE

Abstract

The mechanization and automation of agriculture is a reality, especially in grain crops such as soya and maize. Horticulture, although demanding more labor, suffers from a lack of machinery compatible with the fragility of vegetables, especially for products consumed fresh. The rural exodus in Brazil, coupled with the growing global demand for food, has been a serious problem for many producers, who find labor prohibitively expensive for certain crops. Currently, the cost of labor alone for harvesting staked tomatoes, produced in the open field, is approximately 14,000 reais per hectare - following the average variation between 2016 and 2023, this cost will be approximately 18,000 reais per hectare in 2028.Recently, some foreign companies have successfully developed robots for autonomous tomato harvesting (still in the early stages of operation). However, they were all developed for protected cultivation in greenhouses (a minority in Brazil) and for the cultivars popular in their particular countries, which differ from those most consumed here. Our goal is to be the first company to develop a robot for autonomous tomato harvesting in the open field ( called Pomobot), focusing on the Salada Longa Vida cultivar (predominant in the Brazilian market). The solution was divided into three modules, which will be developed and tested separately: (1) locating and identifying the level of ripeness of the tomatoes using depth cameras and transfer learning from pre-trained You Only Look Once (YOLO) convolutional neural network models; (2) autonomous movement of the robot around the farm, using a skid-steering drive, powered by a combustion generator and guided by the fusion of inertial sensors, wheel odometry and 2D LIDAR combined with the Simultaneous Location and Mapping (SLAM) technique; (3) harvesting and deposition of the tomatoes using robotic arms and soft grippers combined with suction cups.Development will follow an iterative flow that will alternate tests in the simulated world and in the physical world, in a controlled environment (engineering laboratory) and in an operating environment (open field). The aim is to reduce on-site development as much as possible, while ensuring the independent functioning of the modules and their correct integration. We will use Robot Operating System (ROS) as a framework and Gazebo for the simulations, both on a Linux operating system since Pomobot has no critical real-time constraints. At the end of the project, our goal is to deliver a robot capable of autonomously harvesting at least one row of a tomato plantation in an open field, even at sub-productive speed. (AU)