Advanced search
Start date

Development of a Simulation-Based Neural Network Training Pipeline for Improved Navigation of Agricultural Robots through the TerraSentia platform in partnership with the University of Illinois (Urbana-Champaign)

Grant number: 23/15926-8
Support Opportunities:Scholarships abroad - Research Internship - Scientific Initiation
Effective date (Start): April 21, 2024
Effective date (End): July 24, 2024
Field of knowledge:Agronomical Sciences - Agricultural Engineering
Principal Investigator:Marcelo Becker
Grantee:Felipe Andrade Garcia Tommaselli
Supervisor: Girish Chowdhary
Host Institution: Escola de Engenharia de São Carlos (EESC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Research place: University of Illinois at Urbana-Champaign, United States  
Associated to the scholarship:22/08330-9 - Navegação de robôs terrestres em culturas agrícolas utilizando redes neurais em dados LiDAR, BP.IC


The growth in the application and dissemination of artificial intelligence techniques, especially the so-called Neural Networks, also impacts the automation of the agricultural environment. In this context, it is worth highlighting the importance of autonomous robots for rural applications, which can use AI to improve the effectiveness of their results.However, the latent difficulty of carrying out field tests due to logistical issues makes it challenging to obtain massive data. Subsequently, this data must go through a manual labelling process for supervised learning applications. As a result, this exhaustive process of obtaining and annotating data discourages the application of neural networks to the explained context.Having that said, computer simulation environments emerge as an alternative solution to facilitate algorithm testing on the platform and obtaining datasets. This simulation framework enables generating massive datasets containing multiple element variations and countless pre-programmed tests. However, an effective structure is needed for converting the simulation data into a neural network input.This challenge for finding good enough labelled data was noticed during the first-year research of the candidate, when it was necessary to classify 4,000 images manually. In addition, the DASLAB's researchers are also struggling to generate consistent data for the current AI implementations. The Distributed Autonomous Systems Laboratory (DASLAB), led by Dr. Girish Chowdhary, primarily aims to develop highly autonomous mobile robots capable of handling challenging and unpredictable outdoor environments. Whereas the Mobile Robotics Laboratory, under Dr. Becker's supervision in Brazil, shares similar goals, especially in creating fully autonomous systems for agriculture. Therefore, the exchange internship aims to consolidate the great synergy that both institutions can offer. It is worth noting that the University of Illinois at Urbana-Champaign is highly recognized internationally, ranking 12th among public universities and 35th in the United States according to the 2023-24 U.S. News & World Report's America's Best College. Beyond that, according to the Times Higher Education, UIUC is ranked as the 48th worldwide university . Therefore, it is clear that UIUC and DASLAB have the necessary resources and infrastructure to not only support the proposed research, but also to increment the goals and possibilities of this work in the theoretical and practical field. Moreover, this project has the potential to foster collaboration between the LabRoM and DASLAB teams, as well as create additional partnership opportunities with both universities in the future.During the research internship, narrowing down the project's focus, the primary aim is to develop a neural network training pipeline with simulation, validated with subsequent real-life tests. To achieve this, the candidate will collaborate with the DASLAB team, gaining insights into cutting-edge tools used in autonomous robotics development. Furthermore, the candidate will gain a deeper understanding of the TerraSentia platform from EarthSense, a company founded by DASLAB members. Beyond that, the simulation pipeline described can exponentially enhance artificial intelligence works using low resources for training, thus consolidating the next generations of DASLAB AI projects in terms of productivity and reliability. This newfound knowledge is vital for implementing the candidate's crop-follow LiDAR research with neural networks using the same TerraSentia platform, the central goal of his FAPESP project. In addition, the current DASLAB's work-in-progress project with this simulation for AI training aims specifically at a Perception AI-based system, similar to the candidate's research.

News published in Agência FAPESP Newsletter about the scholarship:
Articles published in other media outlets (0 total):
More itemsLess items

Please report errors in scientific publications list using this form.