Using Reinforcement Learning with LiDAR data for Autonomous Navigation

• Grant number: 24/09535-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2024
• End date: August 31, 2026
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Marcelo Becker
• Grantee: Ana Luiza Soares Mineiro Rocha
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated scholarship(s): 25/00527-6 - Federated Reinforcement Learning applied to Autonomous Navigation, BE.EP.IC

Abstract

The project proposes the implementation of an autonomous navigation system for ground robots in agricultural fields, specifically, under the canopy of crops. In this regard, for the robot to navigate between crop rows, comprehensive environmental data acquisition is necessary. Therefore, in this project, the Light Detection and Ranging (LiDAR) sensor will be used, as it is a traditional and reliable sensor in canopy applications, unlike, for example, the Global Positioning System (GPS), which has low accuracy in these environments. From the obtained data, they will be used as input to the navigation system. This system will be prototyped using Reinforcement Learning (RL), being a fundamental part of the research. One of the reasons for choosing RL is its ability to abstract the need for a prior data set, as the RL system learns by trial and error, interacting directly with the environment, thus bypassing the challenges associated with data collection in the agricultural sector. Finally, the navigation approach using LiDAR and RL data proposed in this document will be implemented on the TerraSentia robot, a small ground robot ideal for navigating between crop rows.