RINOBOT-JAGUAR's joint team paper

Ana Sophia C. A. V. Boas, André Luiz C. de Oliveira, Bruna C. Siqueira, Daniel K. Almeida, Félix O. Miranda, Lucas C. V. C. Chaves, Luiz Miguel B. Silva, Marcelo R. de Souza Filho, Marcos V. da Silva, Maria Alice P. Siqueira, Pedro de A. B. Bittencourt, Samuel A. Pereira, Victória T. F. Yamashita, Ana Carolina A. S. Oliveira, Ana Clara C. Genuel, Ana Júlia R. M., Atan de A. Cardoso, Carolina C. G. de Vasconcellos, Diego A. Carvalho, Helton Rodrigo de S. Sereno, Gabriel C. G. de Vasconcellos, Gabriel E. Cortes, Marina Lucia de O. da Silva

University Campus, Federal University of Juiz de Fora (Universidade Federal de Juiz de Fora), St. José Lourenço Kelmer, Juiz de Fora, MG, 36036-900, Brazil; Volta Redonda Campus, Federal Institute of Rio de Janeiro (Instituto Federal do Rio de Janeiro), Rua Antônio Barreiros, 212, Nossa Senhora das Graças, Volta Redonda - RJ, 27215-350, Brasil

www.equipejaguar.com.br

Abstract Team Description Paper of the brazilian joint team Rinobot-Jaguar

Keywords: TDP · Ball Detection· CNN

1 Team Information

Our team is formed by the union of two brazilian teams: the Rinobot Team and the Jaguar. The Rinobot Team was founded in 2016. Its Headquarters stand at the Federal University of Juiz de Fora (UFJF), located in the city of Juiz de Fora MG in Brazil, their the contact information is as follows: [email protected]. It is constituted by more than 60 participants including the captain, Felix Miranda, All of which study Electrical Engineering, Mechanical Engineering, Computer Science, Exact Sciences, Communication, and others, under the guidance of professors from the Electrical Engineering department and a laboratory technician.

The team is divided into 6 areas: Management (responsible for planning, marketing, disclosure, finance, and collection of results), and the other 5 areas are their competition categories. They are: Very Small Size Soccer (VSSS), Line Follower, Lego Sumo, Mini Sumo and Standard Platform League (SPL).

The SPL category is made up of 9 members, including leader Samuel Abreu. These members are distributed among the Exact Sciences, Computer Sciences, Electrical Engineering, and Information Systems courses, all of this at undergraduate level.

Jaguar was founded in 2012. It has its headquarters at the Federal Institute of Rio de Janeiro (IFRJ) on the campus located in the city of Volta Redonda RJ in Brazil. Their contact email is: [email protected]. It is constituted of 23 participants including the captain, Helton Sereno. Excluding him, who is a professor at the IFRJ, graduated in Mechanical Engineering at the Catholic University of Petrópolis (UCP), the other members are studying Technician in Industrial Automation.

The team presents the SPL category as its main area of activity, with a total of 10 active members, all studying Technician in Industrial Automation at undergraduate level.

2 Code Usage

2.1 Vision

All of the vision's systems were developed and maintained by us. It uses OpenCV as a base and now has an AI to identify the ball.

2.2 Motion and Strategy

For the transmission of decisions throughout the game, to the NAO robots in the field and control of the teams on them, the game controller software GameController is used. This software replaces the role of the human referee and offers Human-Machine interface to the teams. It was developed by the German team B-Human and was designated as the official game controller in the matches of the SPL competitions, such as RoboCup. For future competitions we intend to implement a wi-fi communication system between the players to increase the precision of the localization.

Motion and Sonar systems were made by rUNSWift in 2014 and we have been using them since 2019. The team is divided by roles, and, currently we have a goalkeeper and four field players with no distinction among them. Most of the moveset was developed by rUNSWift and encapsulated in a library of easier implementation on our codebase.

3 Own Contribution

3.1 CNN Ball Detection

Since 2018, the Rinobot and Jaguar teams have been using a haar cascade classifier to detect the black and white ball. While this method is considerably fast, it has been found to be unreliable due to the high number of false positives it produces. In order to improve the accuracy of ball detection, the teams have decided to switch to using Convolutional Neural Networks (CNNs) as their classifier.

The new approach consists of two steps. Firstly, the teams use the haar cascade classifier to identify candidate regions of interest in the image. Then, these candidates are fed into a CNN for further analysis. The current implementation uses the tiny-YOLO object detector, which has shown promising results in terms of accuracy and speed.

The teams' CNN is structured with 24 layers, and takes a 32x32x3 input. This design allows for the network to analyze small regions of the image in detail, which can help to reduce false positives and improve overall accuracy. By combining the strengths of both the haar cascade classifier and CNNs, the teams are able to achieve a high level of accuracy while maintaining real-time performance.

The use of CNNs represents a significant improvement over the previous haar cascade classifier approach, as it allows for more sophisticated image analysis and better discrimination between true and false positives.

3.2 Object Detection

The issue with the robots constantly committing fouls during their performances was due to their lack of obstacle detection capabilities. They would focus solely on pursuing the ball without taking into account any obstacles in their path, which often resulted in collisions and fouls.

To address this issue, we implemented obstacle detection through the use of sonar sensors. Currently, we process the data from the two sonars on the robot and simulate a third based on the codebase from the runSwift team. This allows us to obtain more precise information about the location and proximity of obstacles, enabling the robot to determine which side it should move towards to avoid colliding with them.

Overall, this solution has helped to significantly reduce the number of fouls committed by the robots during their performances, resulting in a smoother and more efficient gameplay experience.

4 Past History

Unfortunately, we had no participation in RoboCup competitions since 2018. There is a table that contains the results of the games in RoboCup 2018 which we participated:

Results of Rinobot games in RoboCup 2018

Furthermore, in 2017, the Rinobot Team got the first place in the Latin American Robotics Competition (LARC) on SPL category, which occurred in Curitiba, Paraná, Brazil, and, in 2019, the Rinobot Team got the second place in the same competition, which occurred in João Pessoa, Paraíba, Brazil.

As team Jaguar, we have not yet participated in RoboCup, but we have been participating in LARC since 2016. Our results are presented in the following table:

Ranking of Jaguar in LARC

5 Impact

Robotics is one of the fastest growing areas in the world. New technologies arise at all times and it is undeniable that the future will have the increasingly striking presence of robots in our daily lives. The Rinobot-Jaguar Team, armed with a feeling of growth and innovation, seeks through research and testing, always improve and optimize game strategies.

The Robocup championship, worldwide, has the potential to give more prestige and visibility to the team's efforts, encouraging companies and potential employees to sponsor the team, enabling the continuity of the project.

It will be extremely important for the Brazilian teams that make up the team, because contact with more advanced technologies would contribute greatly to the educational development of the team, in addition to encouraging and strengthening research in this area, in national territory, and being able to put into practice everything we have developed so far.

At Rinobot, the team places a strong emphasis on community outreach. They regularly visit local schools to showcase their robots and introduce children and teenagers to the exciting world of robotics. This is especially important as robotics is still relatively underexplored in basic education, and the team believes in inspiring young minds to pursue careers in this field.

Overall, both the Rinobot and the Jaguar teams are dedicated to advancing the field of robotics through their passion for innovation and their commitment to community outreach. The team's participation in the Robocup is a key component of their mission, as it provides a platform for showcasing their work and advancing the field as a whole.

6 Other

The Rinobot and Jaguar teams are currently collaborating on a project to migrate and integrate the code for the NAO v4 and NAO v6 robots. This involves combining the existing code from Rinobot, developed to the NAO v4 with the code developed by Jaguar for the NAO v6. In addition to this project, the joint team is also working on implementing an autonomous localization system for the robots.

References

1. Wang, X., Girshick, R., Gupta, A., He, K. (2020). Non-local neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 7794-7803).
2. rUNSWift 2014 Code Release, https://github.com/UNSWComputing/rUNSWift-2014-release. Last accessed 14 Feb 2023