Preliminary Validation of an Adaptive Prosthesis Wrist-Hand Control Strategy

Preliminary Validation of an Adaptive Prosthesis Wrist-Hand Control Strategy 1st Enrica Stefanelli CREO Lab, UCBM [email protected] 2nd Francesca Cordella CREO Lab, UCBM [email protected] 3rd Salvatore Li Gioi CREO Lab, UCBM [email protected] 4th Gianmarco Cirelli CREO Lab, UCBM [email protected] 5th Daniele D’Accolti BioRobotics Institute Department of Excellence in Robotics & AI [email protected] 6th Christian Cipriani BioRobotics Institute Department of Excellence in Robotics & AI [email protected] 7th Loredana Zollo CREO Lab, UCBM [email protected] Abstract—This work presents the preliminary validation of an adaptive control strategy for hand-wrist prosthesis, designed to coordinate wrist motion with grasping. The proposed adaptive approach modulates wrist rotation speed in real time based on tangential forces and slip detection to improve motion accuracy, and was compared against a constant velocity control strategy. Experimental validation with eight participants showed a significant reduction in errors when using the adaptive modality, demonstrating that adjusting wrist velocity based on hand–object interaction improves performance in dynamic manipulation tasks. Index Terms—Prostheses, Control, Hand-wrist coordination. I. INTRODUCTION The wrist provides stability and coordination for hand function, supporting accurate movements in daily manipulation tasks. Complex activities, such as pouring or rotating objects, require not only accurate hand positioning but also a dynamic adjustment of wrist motion in coordination with grasp force. In humans, this coordination occurs automatically: the Central Nervous System anticipates destabilizing torques and modulates grasp forces to ensure stability [1]. However, reproducing such anticipatory and adaptive mechanisms in prosthetic systems remains challenging. Current devices rarely implement strategies where wrist rotation dynamics adapt in response to unexpected events during manipulation, such as object slip, which should elicit coordinated reactions not only in grasp force but also in wrist motion. Research efforts in this field have mainly followed two directions. On the one hand, several works have applied machine learning techniques to predict wrist kinematics from bio-signals or other joint data, achieving high accuracy in controlled conditions [2]. On the other hand, alternative approaches have aimed at real-time feasibility, for example regulating wrist rotation based on trunk kinematics [3] or implementing semi-passive wrists with variable stiffness to improve grasp stability during manipulation [4]. While promising, these studies typically addressed either wrist motion or grasp stability in isolation, without explicitly coupling the two. This work was supported INAIL within the BioInterNect (CUP: E57G23000280005) and No Problem (CUP: E57G23000270005) projects. Fig. 1. Proposed adaptive prosthesis hand-wrist control strategy. This gap suggests that the lack of functional coordination between wrist and hand may represent a key limitation in current prosthetic control strategies. Building on this hypothesis, we present a preliminary validation of a novel adaptive wristhand control strategy that explicitly couples wrist rotation with adaptive grasp modulation. By aligning prosthetic control more closely with the natural coordination principles observed in humans, the proposed approach is expected to enhance task performance in complex manipulation tasks. II. MATERIAL AND METHODS The proposed control strategy was grounded on the analysis of human kinematics and grasping forces presented in [5]. Three main factors guided the design: first, a sequential approach was implemented, i.e., the wrist and the hand were separately actuated to reproduce the timing observed in natural movements. Second, the average initial and maximum wrist velocities measured in humans (20◦/s and 50◦/s, respectively) were adopted as reference values for the control design. Lastly, human studies indicated that variations in tangential forces were correlated with wrist motion; more specifically, wrist velocity was reduced in the presence of slip events or significant changes in tangential force to preserve stability and ensure successful task execution. The proposed control strategy is illustrated in Figure 1. The hand was actuated through a force-based control law with an internal position loop and slip detection, in which the fingers were modeled as a virtual finger opposing the thumb. 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629852 213 Fig. 2. Example of the pouring task: t1–t3 indicate finger contacts, s1–s3 slip events; the wrist input shows variations linked to slip or tangential force. A position error exkwas continuously computed to track the desired finger displacement, and a slip correction term (esk) was added to respond to potential slippage. In particular, the balance between tangential and normal forces, up to a corrective factor representing the static friction coefficient, defined a stable grasp, while any excess of force beyond this threshold indicated the onset of slip. The wrist was controlled as a velocity-regulated system, where the reference rotation speed (VR) was modulated in real time based on the forces sensed at the fingertips: it decreased in the presence of slip events or significant variations in tangential forces, and remained constant when grasp conditions were stable. All relevant equations are reported in the block diagram, with further details provided in [5]. Eight healthy participants were asked to use the IH2 Azzurra prosthetic hand and the Ottobock wrist rotator. The hand was equipped with fingertip force sensors on the thumb, index, and middle fingers to measure normal and tangential forces. Muscle activity was recorded through Ottobock sensors placed on the forearm, and a non-linear regression classifier was used to decode user intentions. Participants were instructed to execute five hand and wrist gestures (power grasp, open hand, rest, pronation, supination) to both train the classifier and familiarise with the system. The proposed adaptive strategy (Adaptive), which adjusts wrist velocity in real time according to tangential force variations and slip events, was compared against a baseline constant-velocity strategy (Standard), where the wrist rotates at a fixed speed regardless of fingertip force feedback. Participants completed two functional tasks designed to assess prosthesis performance under precision and temporal constraints: T1) pouring fixed amounts (30 grams) of sugar (S) and yogurt (Y) into a bowl, and T2) inserting as many objects as possible into the box within two minutes. Performance was evaluated using two metrics: the mean absolute error (MAEgrams) of poured contents relative to the target in T1 and the number of blocks successfully inserted (n(blocks)) in T2. Statistical comparisons between the adaptive and standard conditions were conducted (Wilcoxon signed-rank test, p= 0.05). III. RESULTS AND DISCUSSIONS Figure 2 shows an example of the pouring task, highlighting the touch and slip events along with the corresponding wrist response due to slip events and tangential force variations. Figure 3 shows that in T1 the adaptive modality resulted in lower MAE values (10.25 ±9.96 g for S and 7.92 ±11.94 g Fig. 3. Performance comparison between Adaptive and Standard control in block insertion (n(blocks)) and pouring tasks (M AEgrams). for Y) than the standard approach (16.6±10.63 g for S and 20.71±22.45 g for Y), with statistically significant differences (p<0.01 for S, p= 0.01 for Y). These results highlight that real-time modulation of wrist velocity in response to tangential forces and slip events allowed participants to better control prosthetic hand orientation, enabling more accurate and stable pouring. Conversely, the number of successfully inserted items in T2 was higher with the adaptive control (3.93 ±2.26) compared to the standard one (3.00 ±1.89), although this difference was not statistically significant. This indicates that, while the adaptive strategy may assist fine motor control, performance in object alignment and insertion was still limited by factors not directly influenced by wrist velocity modulation, such as prosthesis size and cognitive effort. Overall, these findings show that adaptive control enhances accuracy in object interaction tasks by leveraging natural hand–wrist coordination. IV. CONCLUSIONS This paper demonstrated that the proposed adaptive control strategy significantly improved accuracy in dynamic tasks, reducing error compared to the constant velocity approach. However, no substantial differences emerged in alignment and insertion activities, suggesting that other factors may influence performance. Overall, adaptivity appeared promising in supporting fine and accurate manipulations. Future research will be devoted to the assessment of the cognitive load and potential presence of compensatory movements. REFERENCES [1] H. Ehrsson, A. Fagergren, R. Johansson, and H. Forssberg, “Evidence for the involvement of the posterior parietal cortex in coordination of fingertip forces for grasp stability in manipulation.,” Journal of neurophysiology, vol. 90 5, pp. 2978–86, 2003. [2] S. El-Khoury, I. Batzianoulis, C. W. Antuvan, S. Contu, L. Masia, S. Micera, and A. Billard, “Emg-based learning approach for estimating wrist motion,” 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 6732–6735, 2015. [3] M. Legrand, N. Jarrass´ e, et al., “A closed-loop and ergonomic control for prosthetic wrist rotation,” in 2020 IEEE International Conference on Robotics and Automation (ICRA), pp. 2763–2769, IEEE, 2020. [4] L. Cappello, D. D’Accolti, M. Gherardini, M. Controzzi, and C. Cipriani, “A 2-degree-of-freedom quasi-passive prosthetic wrist with two levels of compliance,” IEEE Robotics and Automation Letters, vol. 8, no. 3, pp. 1231–1238, 2022. [5] E. Stefanelli, M. Lapresa, F. Cordella, D. D’Accolti, C. Cipriani, and L. Zollo, “A hand-wrist control strategy based on human upper limb kinematics,” in 2024 10th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob), pp. 1029–1034, IEEE, 2024. 214