Using Reinforcement Learning with camera data for Autonomous Navigation

• Grant number: 24/09442-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2024
• End date: August 31, 2025
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Marcelo Becker
• Grantee: Arthur Pompeu dos Santos Rocha
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

The project involves the implementation of an algorithm on the TerraSentia robot for terrestrial navigation control in agricultural fields, which are remote areas with numerous terrain irregularities. In these locations, it is essential to obtain comprehensive environmental data to ensure better navigation for the robot. Thus, this research will utilize camera data, which are effective in obtaining high-precision information and features of the environment, aiding in decision-making and enabling the robot to semantically discern obstacles and routes in agricultural environments. With this data, the project proposes the study and implementation of a machine learning method called Deep Reinforcement Learning for TerraSentia navigation, resulting in the creation of adaptive and safe paths in plantations. In this method, the agent learns to make optimal decisions by interacting with the environment dynamically and receiving rewards in exchange for its actions. In this way, an algorithm will be developed to maximize the accumulated reward over time, allowing the robot to navigate precisely between two rows of crops.