05/27/2026
TriSaFe-Trans is a safety-aware multimodal intent recognition pipeline for assistive robotics, developed in the Mecharithm Lab at Saint Louis University.
The paper fuses EEG (brain), EMG (muscle), and eye-tracking signals through a lightweight tri-modal transformer with reliability-gated attention, then applies a safety policy layer that decides whether the robot should act. The pipeline is demonstrated on a Kinova Gen3 arm performing five everyday tasks: reaching for a clock, picking up a dropped bottle, toggling a fan, adjusting a plant, and waving.
Accepted for oral presentation at BioRob 2026, Edmonton.
đ Code: https://github.com/madibabaiasl/SafeIntentDetectionPaper
đ Wiki: https://github.com/madibabaiasl/SafeIntentDetectionPaper/wiki
Authors: Tipu Sultan, Kody Cool, Guangping Liu, Gajapriya Tamilselvan, and Madi Babaiasl
Supported by SLUâs Research Institute, the Clare Boothe Luce Foundation, and SLUâs AEME Department.