Brain-Computer Interfaces (BCI26)

• Shriram Ganapathy, Indian Institute of Science (IISc)
• Sheng-Yu Peng, National Yang Ming Chiao Tung University
• Rupesh Chillale, Ahmedabad University

The BCI (Brain-Computer Interface) work group focuses on science-driven experiments using non-invasive neural recordings. Our current efforts centre on collecting EEG data from human subjects and using it to study motor control, movement intention, and cross-session and cross-subject variability. We aim to develop closed-loop systems where EEG signals are used to control robotic finger movements, enabling precise decoding of movement intention in real time.

A key direction is the development of EEG foundation models that can generalize across sessions and subjects, reducing calibration time and improving robustness in practical BCI applications. Through these studies, we seek to build principled models and algorithms that can later translate into more advanced BCI systems for assistive technologies and neurorehabilitation.

In conjunction with the NIC group, we will also develop neuromorphic circuits for BCI that sample neural signals without conventional ADCs, using event-driven and non-uniform sampling strategies. These approaches aim to reduce power, data rate, and system complexity while preserving behaviorally relevant information, paving the way for more efficient and practical BCI hardware.

• EEG-based Motor Intention Decoding and Closed-loop Control for Robotic finger/hand control using real-time EEG.
• EEG Foundation Models and Robust Decoding for Cross-session and cross-subject generalization.
• Neuromorphic Sampling and Circuits for BCI for ADC-less, event-driven / non-uniform sampling in collaboration with NIC.
• Experimental Neuroscience for BCI for Human EEG data collection, protocol design, and behavioural studies.

• Motor intention decoding and closed-loop robotic finger control using EEG.
• Design and evaluation of four different analog-to-spike converters as a continuation of the BNEW25 work.
• Level-crossing ADC circuits, including Python/MATLAB modelling to reduce sampling and conversion times efficiently and explore SVR or super-resolution neural networks for high-resolution reconstruction from fewer samples.
• EEG Foundation Models for Cross-Subject and Cross-Session Variability.
• Collaborative projects with OnSor.
• More Project Ideas will be added to this list soon.