AI Rule Compliance Policy Announcement [RoboCup2025]

Dear teams,

We are sharing an important update regarding the use of AI tools and compliance with Rule 4.1 in this year’s competition.

:hammer_and_wrench: Acknowledging Our Responsibility

This year, the Committee conducted a detailed review of team documentation prior to the competition, aiming to ensure fairness and consistency in the application of Rule 4.1. However, we found that some teams interpreted the rule differently from what was originally intended. The original wording may not have been sufficiently clear, and while we issued several clarifications via the forum, they were not always consistent — potentially causing further confusion, particularly for teams that were trying to comply in good faith.

We deeply regret the confusion this has caused and feel a strong sense of responsibility. We are sincerely grateful for your continued efforts and patience despite the uncertainty. Many teams have put significant time and energy into their work — often under unclear guidance — and we truly admire your resilience and commitment to learning.

Our goal remains to create a fair and transparent competition where teams can showcase their learning, apply their creativity responsibly, and be recognized for their contributions. In this spirit, we have decided not to penalize teams, but instead to positively recognize those whose work clearly aligns with both the rules and the educational mission of RoboCupJunior.

:star2: A Positive Approach: Bonus Instead of Penalty

To fairly account for compliance and effort, we are applying a field score bonus ranging from 0% to 10%, depending on how each team’s AI-related implementation aligns with Rule 4.1 and the level of originality and technical contribution demonstrated.

These categories reflect not only rule interpretation, but also the educational challenge, creativity, and depth of understanding shown by the teams.

In addition to the field score bonus, Rule 4.1 compliance and the team’s AI implementation approach may also be considered during the selection of Side Awards and other forms of recognition. Teams that demonstrate clear alignment with the spirit and intent of the rule — especially through creative and educationally meaningful approaches — may be positively acknowledged beyond scoring.

:mag: AI Compliance Buckets and Bonuses

Bucket Description Bonus
1. AI not used at all The robot logic is based entirely on sensors and traditional programming. No image processing or AI components are used. 10%
2. Traditional computer vision(e.g., OpenCV-based pattern matching, color filtering) The team uses classic image processing techniques such as contour detection, color thresholding, or template matching via tools like OpenCV. No machine learning or AI model is used, but camera data is processed with rule-based algorithms. 10%
3. Custom AI model fully developed from scratch The team designed and implemented their own neural network architecture, selected or implemented loss functions, and performed the full training process using a self-labeled dataset. This reflects deep understanding and significant technical contribution. 10%
4. Significant modification of a known model (architecture/loss function/tensors changed) The team reused a known model architecture (e.g., YOLO, MobileNet), but modified internal elements such as the number of layers, neurons, tensors, or loss functions, and trained it with their own dataset. This involves a substantial level of understanding and customization. 10%
5. Standard architecture without pre-trained weights, or with significant technical contribution A well-known model architecture is used without changes, and trained entirely with the team’s own labeled dataset. Alternatively, the team used a standard architecture and training pipeline, but made a significant technical contribution — such as identifying bugs, contributing to upstream improvements, or developing custom tooling related to the AI framework. 7%
6. Fine-tuning a pre-trained model with own dataset (transfer learning) The team used a pre-trained AI model and further trained it using their own labeled images. The core architecture and training setup remain mostly unchanged. 3%
7. Minor modifications to a pre-trained model A pre-trained model is used as-is, with only light adjustments like confidence thresholds or detection parameters. No retraining or new labeling was done. 0%

:envelope_with_arrow: How to Check Your Team’s Classification

Each team has received a personalized email including:

  • Their assigned AI Compliance Bucket
  • The corresponding field score bonus
  • A link to submit a review request if needed

Please check your team’s registered contact email for these details.


:clock3: How to Submit a Review Request

If you believe your classification is inaccurate or would benefit from further clarification, you are welcome and encouraged to request a review.

To do so:

  1. Log in to your Personal Page on the CMS
  2. Find the form titled “AI Compliance Bucket Review Request”
  3. Indicate the bucket you believe best matches your team
  4. Provide supporting explanation or documentation

Deadline: July 14, 23:59:59 UTC

The Committee will review all requests carefully. If needed, on-site interviews may be held during the competition to finalize each team’s classification.


:speech_balloon: Final Note

Your feedback and engagement help us improve not only this year’s competition, but also the clarity and fairness of the rules in future years. Thank you for being part of this learning journey.

We sincerely appreciate your understanding and the tremendous effort you’ve invested. We look forward to seeing your work at the competition.

Warm regards,
RoboCupJunior Rescue Committee


Note:
Throughout this post, references to “AI” specifically refer to existing AI tools, such as pre-trained models, machine learning frameworks, or dedicated AI hardware. While we understand that other approaches like finite state machines (FSMs) can also be seen as AI in a broader sense, our classification here focuses only on the use of external AI-based software or hardware systems.

1 Like

Dear the Committee,

Thank you for the clear and thoughtful announcement explaining how the situation will be addressed following the confusions regarding the AI use. The proposed solutions not only ensures positive experiences for all teams but also offer valuable opportunities for them to gain a deeper understanding of the AI use rules established by the committee, preparing them for future events.

Warmly Amy