We understand that, according to the rules and the latest communications, using a pre-trained model without making significant modifications is not considered development and therefore is not allowed.
Our question is the following:
If a team uses Edge Impulse as a starting point, but then exports the model, modifies it, and is also able to clearly explain how the model works, what changes were made, and why — would this be allowed under the rules?
Or is the use of Edge Impulse completely prohibited, even if further modifications are made?
Thank you for your question regarding the use of Edge Impulse and model modifications.
According to the current rules and clarifications, simply using a pre-trained model without substantial changes is not considered sufficient development and is therefore not allowed as meeting the development requirements.
However, if a team uses Edge Impulse as a starting point but then exports the model and makes significant modifications — such as altering the architecture, changing loss functions, adding or removing layers, or other meaningful customizations — and is able to clearly document and explain these changes and their rationale, this can be considered valid development work.
In such cases, the team’s approach would be reviewed carefully by the committee to confirm that the modifications meet the criteria for substantial technical contribution as outlined in the rules.
To summarize:
Using Edge Impulse purely as a tool for training a standard pre-built model without changes is not allowed.
Using Edge Impulse as a base but performing significant, well-documented modifications may be allowed and considered under a higher compliance category.
The above is an explanation regarding the interpretation of the rules.
For details about how they will be applied at RoboCup 2025 in Salvador, please refer to the following article:
We encourage teams to provide detailed documentation on their development process to facilitate evaluation.