Neural Engineering
JuiHsuan Wong
Undergraduate Researcher
Earlham College, United States
Laura McGann
PhD Candidate
Case Western Reserve University, United States
Rachel Jakes (she/her/hers)
PhD Candidate
Case Western Reserve University, United States
Dustin Tyler
Kent H. Smith II Professor of Biomedical Engineering
Case Western Reserve University, United States
The sense of touch is an essential aspect of human perception, facilitating interaction with our surroundings [1]. As humans increasingly interface with environments via robotic devices, including prostheses and other robotic manipulators, the ability to mimic the intricate and delicate movements of a human hand becomes increasingly desirable [2]. Despite the development of high degree-of-freedom (DOF) robotic devices, achieving precise and intuitive control remains a significant challenge, particularly due to a lack of sensory feedback [3], [4].
Natively, the sense of touch is elicited by two classes of mechanoreceptors: slowly-adapting (SA), which detects sustained pressure on the skin and most directly mimics the stimulus, and rapidly-adapting (RA), which best detects change. Current force-sensing techniques primarily focus on slowly-adapting analogs, limiting the captured information. To address this issue, our research investigates the value of two robotic sensors: Force-Sensing Resistors (FSR) and piezoelectric materials.
While FSRs are commonly employed to measure force directly, similar to an SA mechanoreceptor, their larger form factor limits the application area and measurement reliability across their surface area. Piezoelectric sensors present a smaller force-transducing alternative with the potential to capture RA-analogue information and enrich the tactile feedback relayed to the user (Figure 1).
To evaluate the sensitivity of the FSR and piezoelectric sensors across their receptive fields, we analyzed sensor performance under a range of applied weights across different contact locations. By understanding the sensor sensitivity profiles, we can determine how to apply these sensors for closed-loop control of neuroprostheses and other robotic devices.
Our study aims to assess the impact of contact location on force sensor performance for both the slowly-adapting sensor (FSR) and the rapidly-adapting sensor (piezoelectric). We tested a variety of location-weight combinations for each sensor, five trials per condition. Locations were spaced 0.25 cm apart along the central longitudinal axis of each sensor, yielding five and three distinct locations for the FSR and piezoelectric sensors, respectively, due to their different sizes.
Masses were applied to specific points either via Von Frey filaments for lower weights or a spring-suspended platform for larger masses. The platform apparatus was designed using Onshape and 3D-printed. A pin protrudes from the bottom of the platform to allow precise force application to specific sensor locations when masses are placed on the platform. Concentric rings in the platform surface secure various mass sizes.
We obtained the per-location sensor output curves by applying a range of applied weights at each location. Von Frey filaments were applied directly for 4g, 6g, 8g, 10g, 15g, 26g, 30g, 60g, 100g, and 180g masses. The 3D-printed apparatus was used for masses of 200g, 300g, and 1000g. The force applied to the sensor via the spring-mass system was computed as Fsensor = Fplatform+ Fmass - Fsprings, where Fplatform and Fmass are computed using F=m*g and Fsprings = 4*kspring*Δxspring.
The FSR and piezoelectric sensors were tested over 30g-100g and 4g-200g ranges, respectively. Additionally, we integrated the output readings over time for the piezoelectric sensor to account for the output signal peak and width.
Based on the obtained force-output curves (Figure 2), we selected masses 60g and 100g for analysis.
First, we assessed the stability of each sensor output across contact locations. For both sensors, ANOVA and Tukey's HSD tests showed significant differences between at least one pair of the tested locations. The FSR sensor was more sensitive at central locations compared to the first and last locations (Figure 3). At 60g on the FSR, the most distal location (Loc5) returned no output and thus was significantly different from all the other locations (p< 0.001, Fig3a). At 100g, location 5 was significantly different from locations 2-4 (p< 0.01, Fig3b). Overall, only the most distal location had a consistently different sensitivity on the FSR receptive field.
The piezoelectric sensor was most sensitive distally (Loc3). At 60g, only location 3 showed a significant difference (p< 0.001, Fig3c). At 100g, location 3 was still significantly different from locations 1-2 (p< 0.001), but locations 1-2 were also significantly different (p< 0.05, Fig3d). Overall, location had a significant effect on the piezoelectric sensor output, indicating low sensor stability across its receptive field.
Second, we investigated the sensitivity of each sensor to different masses (Figure 4). The FSR exhibited a significant difference between 60g and 100g (p< 0.01), highlighting its ability to discriminate between different masses. The piezoelectric sensor outputs were not statistically different between the two masses.
Discussion: These findings indicate FSRs provide more stable measurements across their receptive field compared to piezoelectrics, suggesting continued FSR use for primary force transduction. However, the location-influenced activation of both sensors still limits their receptive fields to smaller areas than their physical form factors, and additional sensors should be considered in future.
Additionally, FSRs better discriminate between different forces applied. The piezoelectric's low sensitivity to different masses suggests its suitability for tactile submodalities like RA edge detection, saving SA continuous monitoring for FSRs.
This research informs the capabilities and best uses of two sensors for artificial tactile sensation. Applying each sensor appropriately to robotic devices can enrich sensation and perception of environments through human-robot interfaces.
I am sincerely grateful to Dr. Dustin Tyler and the entire lab team, especially Ph.D. candidates Rachal Jakes and Laura McGann, for their invaluable mentorship and guidance throughout this summer. Their support and encouragement have allowed me to contribute to the field of neuroengineering research actively, and I am truly honored to have been a part of their team. Furthermore, thanks to the APT Program for providing the essential resources and opportunities that made this enriching experience possible. I wish to thank the Center of Global Health at Earlham College for generously funding and supporting this project. Their financial support has been crucial in making this research endeavor a reality, and I am deeply grateful for their commitment to advancing scientific exploration and knowledge.
[1] Jenkins, B. A., & Lumpkin, E. A. (2017). Developing a sense of touch. Development, 144(22), 4078–4090. https://doi.org/10.1242/dev.120402
[2]Sihombing, P., Muhammad, R. B., Herriyance, H., & Elviwani, E. (2020). Robotic arm control based on fingers and hand gesture. 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT). https://doi.org/10.1109/mecnit48290.2020.9166592
[3] F. Amirabdollahian et al., “Prevalence of haptic feedback in robot-mediated surgery: a systematic review of literature,” J. Robot. Surg., vol. 12, no. 1, pp. 11–25, 2018, doi: 10.1007/s11701-017-0763-4.
[4] M. K. Chmarra, J. Dankelman, J. J. van den Dobbelsteen, and F.-W. Jansen, “Force feedback and basic laparoscopic skills,” Surg. Endosc., vol. 22, no. 10, pp. 2140–2148, 2008, doi: 10.1007/s00464-008-9937-5