This paper introduces a multi-modal deep learning framework, BioFEN, for predicting human intent and estimating physical effort in real-time using surface electromyography, inertial measurement units, and force sensors. The framework addresses individual variability, fatigue dynamics, and real-time constraints, aiming for seamless human-robot collaboration.
Key findings
BioFEN integrates sEMG, IMUs, and force sensors within a deep learning architecture.
The framework enforces biomechanical plausibility through a constrained loss function.
Achieves real-time inference with accuracy within 10% of ground truth and latency below 50ms.
Includes an online adaptation mechanism for personalization without extensive calibration.
Limitations & open questions
The framework's performance in diverse real-world scenarios needs further validation.
The impact of sensor noise and displacement on long-term accuracy is yet to be fully assessed.