Research at CY Cergy-Paris Université

I was selected as a research intern under the "France Excellence Bourses" 2025, a French government scholarship program for Japanese doctoral students. I conducted collaborative research with Prof. Alexandre Pitti and Prof. Mathias Quoy at the Équipe Traitement de l'Information et Systèmes (Information Processing and Systems Team), CY Cergy-Paris Université.

In this research, we proposed a model that integrates predictive coding and class embedding, enabling trajectory generation, class recognition, and meta-recognition of recognition confidence in real time on a physical robot. A key contribution of this work is that these three functions are all realized within a single RNN model (not a combination of multiple modules), meaning that no task-specific retraining or separate computation is required. All three emerge from a unified process of prediction-error minimization, which keeps the model computationally lightweight and scalable.

While models such as PV-RNN can already handle generation and recognition within a single framework, to our knowledge, no existing single-model approach explicitly performs class identification along with meta-estimation of recognition confidence. Although large-scale models combining multiple modules can achieve this, they lack biological plausibility and are computationally too heavy for real-time operation.

This work has been submitted to IEEE Robotics and Automation Letters (RA-L) and is currently under review. I will share the details here once the paper is officially accepted and published.

For now, although it is still uncertain whether it will be accepted within the first six months of my stay in France, I am glad that we were able to submit the paper during this period.