Dear @erne1 ,
Thank you for your message and for sharing your concerns and sorry for late response.
We understand that the AI-related rules can be difficult to interpret, especially for teams new to machine learning. To clarify, simply switching from one known model architecture (e.g., YOLOv8n) to another (YOLOv8s) and training on your own dataset, without modifying the internal structure such as loss functions, layers, or tensors, generally falls into the category of minor modification or fine-tuning, which may not be sufficient to meet the higher-level technical contribution criteria under this year’s rules.
That said, we recognize that teams new to these technologies may find it challenging to redesign complex components like loss functions, and the competition aims to balance inclusivity with educational value and fairness.
We encourage all teams to carefully review the AI compliance criteria that have been shared now.
Regarding detailed information on this year’s AI rule enforcement and policy, please refer to the email sent to all teams or to the following forum post:
If your team is planning further development or can demonstrate deeper technical understanding or modification—such as changing architecture internals or adding custom layers—you can submit additional documentation to request a review of your category before the competition.
Please be aware that the policies and interpretations explained in that post apply only to RoboCup 2025 Salvador. The AI-related rules and their enforcement for future competitions, including RoboCup 2026, will be revisited and discussed from scratch after this year’s event.
Your feedback, along with that of other teams, is highly valuable for these future discussions. We sincerely appreciate your engagement and encourage you to continue sharing your opinions to help us improve the rules and their clarity going forward.
Best regards,
RoboCupJunior Rescue Committee