Online Adaptation for Myographic Control of Natural Dexterous Hand and Finger Movements

Abstract: One of the most elusive goals in myographic prosthesis control is the ability to reliably decode continuous positions simultaneously across multiple degrees-of-freedom. Goal: To demonstrate dexterous, natural, biomimetic finger and wrist control of the highly advanced robotic Modular Prosthetic Limb. Methods: We combine sequential temporal regression models and reinforcement learning using myographic signals to predict continuous simultaneous predictions of 7 finger and wrist degrees-of-freedom for 9 non-amputee human subjects in a minimally-constrained freeform training process. Results: We demonstrate highly dexterous 7 DoF position-based regression for prosthesis control from EMG signals, with significantly lower error rates than traditional approaches (p < 0.001) and nearly zero prediction response time delay (p < 0.001). Their performance can be continuously improved at any time using our freeform reinforcement process. Significance: We have demonstrated the most dexterous, biomimetic, and natural prosthesis control performance ever obtained from the surface EMG signal. Our reinforcement approach allowed us to abandon standard training protocols and simply allow the subject to move in any desired way while our models adapt. Conclusions: This work redefines the state-of-the-art in myographic decoding in terms of the reliability, responsiveness, and movement complexity available from prosthesis control systems. The present-day emergence and convergence of advanced algorithmic methods, experiment protocols, dexterous robotic prostheses, and sensor modalities represents a unique opportunity to finally realize our ultimate goal of achieving fully restorative natural upper-limb function for amputees.

• The paper presents an innovative online adaptation approach for myographic control of natural dexterous hand and finger movements using sequential temporal regression models and reinforcement learning.
• The research employs temporal convolutional networks (TCN) and LSTM models, demonstrating superior performance over SVR with lower error rates (around 8-10°) and negligible prediction delays in both structured and freeform experiments.
• This study offers a promising framework for achieving naturalistic, responsive prosthetic control with potential for immediate clinical application and future integration with advanced sensor modalities like sonomyography.

Online Adaptation for Myographic Control of Natural Dexterous Hand and Finger Movements

The paper "Online Adaptation for Myographic Control of Natural Dexterous Hand and Finger Movements" by Betthauser et al. presents an innovative approach to the control of prosthetic devices using myographic signals. The focus is on achieving reliable and natural control across multiple degrees of freedom (DoFs), specifically targeting the simultaneous control of seven DoFs encompassing finger and wrist movements.

Summary and Key Findings

The authors employ a combination of sequential temporal regression models and reinforcement learning to predict continuous movements in prosthetic limbs. They utilize electromyographic (EMG) data collected from non-amputee subjects to accomplish this task. The main objective is to enable dexterous, biomimetic control that mimics natural hand and finger movements, thus enhancing the user experience for prosthesis users.

Key technological components include the use of temporal convolutional networks (TCN) and long short-term memory (LSTM) recurrent networks, both capable of processing time series data to predict movements. These models are evaluated against traditional support vector regression (SVR), demonstrating superior performance in terms of lower error rates and negligible prediction response time delays.

Experimental Design

Two experimental paradigms are presented:

• Standard Experiment: A structured approach where subjects perform pre-defined movements in response to cues, allowing the testing of multi-DoF position-based regression performance.
• Freeform Reinforcement Experiment: A more flexible approach where subjects freely perform a variety of movements, enabling the model to adapt in real-time, reflective of a natural use case scenario.

Both paradigms serve to establish the robustness of the proposed models, with TCN and LSTM models outperforming SVR in both angular root-mean-square errors (RMSE) and prediction response time delay, with RMSE around 8° for structured experiments and sub-10° for freeform movements.

Implications and Future Directions

The research highlights the potential for achieving natural, responsive control in myographic prosthesis, significantly contributing to the field of prosthetic development and adaptive control systems. This study opens avenues for immediate translational applications in clinical settings, with a potential future trajectory towards fully functional myoelectric-controlled prostheses for amputees.

Moreover, the paper anticipates that advanced robotic devices such as the Modular Prosthetic Limb will become more accessible as costs decrease, further enhancing the applicability of the research. The convergence of advanced algorithmic methods with versatile sensor modalities like sonomyography could lead to even more sophisticated prosthetic solutions in the future.

Conclusion

The study presents substantial advancements in achieving naturalistic and adaptive prosthetic control using EMG signals. By demonstrating the efficacy of combining sequential models with reinforcement learning, the authors provide a forward-thinking framework for natural upper-limb function restoration. Future work is likely to expand on integrating sonomyographic sensors, offering a broader perspective on multi-modal prosthetic control systems and underscoring the paper's significance in progressing towards fully restorative limb function for amputees.