We have been working in the last years on a project in our lab named "Learning Setplays from Demonstration." The main goal of this project is to provide evidence that we can catch human's intuitive knowledge in a domain and use it to feed a massive dataset for robots to learn how to cooperate to solve problems in this domain.
At this time, we are gathering contributions worldwide to populate the dataset. So we ask your contribution in the following way: 1- Watch some robot soccer games 2- Locate situations where you think you can demonstrate a setplay for a team to perform better in that situation 3-Create your demonstration 4- Send us several demonstrations
To turn it easier for you, we have developed the BahiaRT's Learning Setplays from Demonstration Toolkit that encapsulates in Docker containers all tools you need to watch games and create setplays demonstrations. The following links contain all information you need:
- The BahiaRT's Learning Setplays from Demonstration Toolkit - https://bitbucket.org/bahiart3d/setplaysdataset/src/master/ - Setup instructions: https://bitbucket.org/bahiart3d/setplaysdataset/src/master/README.md - Usage instructions: https://bitbucket.org/bahiart3d/setplaysdataset/src/master/USAGE.md - Video tutorial: https://youtu.be/h_s8rA2IS88
At the end of this project, we will provide: 1- A public massive setplays dataset organized using our fuzzy clustering strategy ready to be used for any soccer team (not only 3D simulation teams) 2- Source code of the dataset organizer and the deep reinforcement learning engine used for teams training. 3- Results published (Ph.D. thesis and several papers).
You can see some earlier publications of this project at the end of this message.
We appreciate your contribution. Anyone who enjoys soccer can contribute. You do not need any experience with robot soccer or 3D Soccer Simulation. I hope you have fun and send us several setplays demonstrations in the following weeks.
If you have any doubts, please send a message to firstname.lastname@example.org.
Earlier Publications: M. A. C. Simões et al., “Generating a dataset for learning setplays from demonstration”, SN Appl. Sci., vol. 3, nº 6, p. 608, jun. 2021, doi: 10.1007/s42452-021-04571-y. M. A. C. Simões, R. M. da Silva, e T. Nogueira, “A Dataset Schema for Cooperative Learning from Demonstration in Multi-robot Systems”, J Intell Robot Syst, vol. 99, nº 3–4, p. 589–608, set. 2020, doi: 10.1007/s10846-019-01123-w. M. A. C. Simoes et al., “Strategy Planner: Enhancements to support better defense and pass strategies within an LfD approach”, in 2020 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC), Ponta Delgada, Portugal, abr. 2020, p. 46–52. doi: 10.1109/ICARSC49921.2020.9096188. M. A. C. Simões e T. Nogueira, “Towards setplays learning in a multiagent robotic soccer team”, in 2018 latin american robotic symposium, 2018 brazilian symposium on robotics (SBR) and 2018 workshop on robotics in education (WRE), nov. 2018, p. 277–282. doi: 10.1109/LARS/SBR/WRE.2018.00058.