Abstract
Bidirectional human-machine interfaces involve commands from the central nervous system to an external device and feedback characterizing device state. Such feedback may be elicited by electrical stimulation of somatosensory nerves, where a task-relevant variable is encoded in stimulation amplitude or frequency. Recently, concurrent modulation in amplitude and frequency (multimodal encoding) was proposed. We hypothesized that feedback with multimodal encoding may effectively be processed by the central nervous system as two independent inputs encoded in amplitude and frequency, respectively, thereby increasing state estimate quality in accordance with maximum-likelihood estimation. Using an adaptation paradigm, we tested this hypothesis during a grasp force matching task where subjects received electrotactile feedback encoding instantaneous force in amplitude, frequency, or both, in addition to their natural force feedback. The results showed that adaptations in grasp force with multimodal encoding could be accurately predicted as the integration of three independent inputs according to maximum-likelihood estimation: amplitude modulated electrotactile feedback, frequency modulated electrotactile feedback, and natural force feedback (r² = 0.73). These findings show that multimodal electrotactile feedback carries an intrinsic advantage for state estimation accuracy with respect to single-variable modulation and suggest that this scheme should be the preferred strategy for bidirectional human-machine interfaces with electrotactile feedback.
Introduction
Modern upper-limb prostheses often lack a robust "return path" for sensory information. While users can send motor commands, they receive minimal feedback regarding the state of the device, such as grip force. This necessitates heavy reliance on visual monitoring, which is cognitively demanding and prevents natural interaction. Electrotactile stimulation—delivering small electrical pulses to the skin—offers a non-invasive method to restore this sense of touch. Traditionally, a single physical variable like force is mapped to either the amplitude (intensity) or frequency (rate) of the stimulation. Recent advancements have proposed "multimodal encoding," where both parameters are modulated simultaneously. This study investigates whether the central nervous system (CNS) treats these parameters as independent information channels and integrates them according to the principles of Maximum Likelihood Estimation (MLE) to improve state estimation.
Materials and Methods
The study involved 14 able-bodied participants and one transradial amputee. The primary task was a grasp-force matching experiment. Participants were asked to reach a target force (15–30% of maximum voluntary contraction) using visual feedback. Once the target was reached, visual feedback was removed, and participants had to maintain or recreate the force using only natural proprioceptive feedback and electrotactile feedback.
Three electrotactile encoding schemes were tested: Amplitude Modulation (AM), Frequency Modulation (FM), and Multimodal (MM). To quantify the integration process, researchers employed an adaptation paradigm. By introducing a subtle "bias" or shift in the feedback-to-force relationship, they could observe how participants adjusted their grip. This allowed for the calculation of the relative "weight" assigned to the artificial feedback versus natural feedback. The MLE framework was then used to predict multimodal performance based on the individual AM and FM data.
Results
The experimental results strongly supported the hypothesis that the CNS treats multimodal electrotactile parameters as independent sensory streams.
Precision and Accuracy
Multimodal encoding significantly outperformed single-parameter modulation. As shown in the force traces and statistical summaries, the MM condition resulted in the lowest plateau force error and the lowest variance in force production.
Specifically, the "Just Noticeable Difference" (JND)—the smallest detectable change in force—was approximately 9% for multimodal feedback, compared to 17% for AM and 19% for FM. This indicates a substantial improvement in sensory resolution.
Statistical Integration (MLE)
The MLE model demonstrated high predictive power. The correlation between the weights predicted by the model and those observed experimentally was r² = 0.73. For the variance of the force estimates, the correlation was r² = 0.67. These findings suggest that the brain performs "sensor fusion" on the AM and FM components of the signal, treating them as independent sources of information about the same underlying variable.
Amputee Validation
The data from the transradial amputee participant closely mirrored the trends observed in the able-bodied group. The distribution of normalized plateau forces for the amputee demonstrates that the ability to integrate artificial tactile feedback remains intact following limb loss.
Discussion and Conclusion
This research provides a theoretical foundation for why multimodal electrotactile feedback is superior to single-parameter modulation. By leveraging the brain's natural capacity for optimal sensory integration (MLE), multimodal encoding effectively increases the signal-to-noise ratio of the feedback. Furthermore, participants reported that multimodal stimulation was more pleasant and felt more efficient than amplitude modulation alone.
While the results are promising for wearable tech, it remains to be seen if the same MLE-based integration holds for intraneural (implanted) electrodes. Future work should also explore the "information bandwidth" of this approach—specifically, whether multiple independent variables can be encoded simultaneously using different parameters without cross-talk interference. In conclusion, multimodal electrotactile feedback represents a significant step forward for bidirectional human-machine interfaces, enhancing the precision and reliability of prosthetic control.
Summary updated