Learning-Based NMPC for uncertainty mitigation for agricultural rover path tracking

• Grant number: 23/17678-1
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: April 01, 2024
• End date: July 31, 2024
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Marcelo Becker
• Grantee: Francisco Affonso Pinto
• Supervisor: Girish Chowdhary
• Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: University of Illinois at Urbana-Champaign, United States
• Associated to the scholarship: 22/03339-8 - LQR control use for robotic navigation control in agricultural fields, BP.IC

Abstract

The project focuses on developing a controller for agricultural rovers like theTerrasentia capable of handling uncertainties in challenging environments. To achievethis, the proposed approach involves employing the nonlinear optimal controller Non-linear Model Predictive Controller (NMPC) to generate suitable control actions forprecise path tracking. Additionally, a deep neural network (DNN) will complementthis approach to address dynamic model uncertainties in complex environments.The architecture allows for real-time unsupervised training to manage uncertaintieseffectively.The implementation will rely on a set of techniques and software tools for eachstage. Initially, CasADi will handle the symbolic construction of the dynamic model,followed by DNN development using PyTorch. This process forms the optimizationproblem, allowing us to utilize IPOPT as a solver. Finally, Robot Operating System(ROS) integration will be employed to connect with other modules in the robot'snavigation system.