Robot-Assisted Training to Improve Proprioception of Wrist

In recent years, robot-assisted training has been shown to significantly improve motor function and proprioception in people with functional disabilities, but the efficiency of proprioceptive acuity was unclear. To characterize the efficiency of joint proprioceptive acuity improvement in space, we designed a robot-assisted ipsilateral joint position matching experiment using the wrist as the study object. We conducted 2-way repeated measures ANOVA on error data before and after training in 12 healthy subjects and mapped the distribution of wrist proprioceptive learning ability in different workspaces. The results showed significant differences in the proprioceptive acuity of the wrist joint in different workspaces and movement directions before and after training in 12 subjects (<inline-formula> <tex-math notation="LaTeX">$\text{p} < 0.01$ </tex-math></inline-formula>), and the proprioceptive acuity of the wrist after training was significantly higher than before training. In addition, the learning ability of wrist proprioceptive acuity showed significant differences in different workspaces and movement directions (Flexion and Extension in habit workspace (HW) (<inline-formula> <tex-math notation="LaTeX">${P}={0}.{037}$ </tex-math></inline-formula>); Flexion and Extension in maximum workspace (MW) (<inline-formula> <tex-math notation="LaTeX">${P}={0}.{016}$ </tex-math></inline-formula>); Flexion in HW and MW (<inline-formula> <tex-math notation="LaTeX">${P}={0}.{043}$ </tex-math></inline-formula>)). Robot-assisted training is beneficial for improving the proprioceptive acuity of the wrist. The learning ability of proprioceptive acuity of joints in different movement directions is independently distributed and influenced by usage habits, which accelerate the improvement of proprioceptive acuity. This research hopes to clinically guide the development of highly effective rehabilitation programs to achieve better recovery and help build patient confidence.

and skin during exercise or rest, including position sensing, motion sensing, and vibration sensing [1].Patients with functional disabilities are often accompanied by proprioceptive deficits that manifest as insensitivity to movement and position information [2], severely affecting their daily lives.Some researchers [3], [4] found that improving proprioceptive acuity helps to improve or even restore the movement function of the limb.Research on the general approach to improving proprioceptive acuity is vital to help restore movement ability in patients with movement dysfunction.
As robotics technology develops, more and more rehabilitation robots are used for limb rehabilitation and proprioceptive evaluation [5], [6].Some studies have found that training induces plasticity in the human motor networks [7], [8].Repeated proprioceptive training [9], [10] and passive limb movement training [11], [12] are helpful for the recovery of proprioceptive and motor functions in stroke survivors.Marini et al. used a robot-aided method to measure proprioceptive function objectively [13].They used position sense to characterize the proprioceptive acuity in the wrist's three degrees of freedom (DoF) in different workspaces.However, they divide the workspace based on the percentage of wrist range of motion (ROM), ignoring the influence of usage habits.
Several studies showed variability in proprioceptive acuity in different movement directions and spaces [13], [14], which may signal that the extent of improvement in joint proprioceptive acuity with training also varies.In studies of limb motor control, Darainy et al. showed that brief periods of reinforced perceptual training have durable effects on the rate and extent of motor learning [15].Wong et al. suggested that passive arm movement increases the extent of motor learning [16].In addition, proprioceptive improvements may be transferred to other functional tasks involving the same joint or limb movement system [11], [17], which means that finding efficient and rapid training methods to improve joint proprioceptive acuity can be beneficial in shortening the rehabilitation time.
In conclusion, most studies have focused on assessing proprioceptive acuity, and training helps improve proprioceptive acuity.However, the factors that influence the efficiency of proprioceptive acuity improvement are unclear.Therefore, to characterize the general pattern of joint proprioceptive acuity learning ability, we propose a research method that reveals the effect of training on joint proprioceptive learning ability through wrist robot-assisted training.The specific objectives are to statistically and analytically determine the impact of usage habits and movement direction on the efficiency of wrist proprioception improvement under the same training conditions.

A. Participants
Twelve right-handed subjects (age 23.9±1.6 years, 5 females, 7 males) recruited from local universities with no history of neuromuscular disease and unfamiliar with the task participated in the study.The handedness of all participants was assessed by the Edinburgh Handedness Inventory [18].The study was approved by the Institutional Review Committee of Shanghai University of Medicine & Health Sciences (protocol number 2019MQY).And each subject signed a consent form.Experiments were carried out at the Laboratory of Shanghai Institute of Rehabilitation Engineering Technology.

B. Wrist Robotic System
The experimental apparatus used in this study was two DoFs wrist robot system (Fig. 1) specifically designed to study human movement rehabilitation [19].The wrist robot was driven by two brushless motors equipped with a high-resolution incremental encoder to record the joints' rotation angle during movement.It can assist the user's wrist to reach the target position accurately to ensure that each training is effective.During the target position matching process, the direct self-control scheme was used to generate an auxiliary force field, which was used to eliminate the frictional force of the drive to reduce the adverse effect of the equipment resistance in the experiment.During the experiment, to exclude the influence of multi-joint compound motion on the experimental results, the subjects could only operate the wrist robot freely in the flexion/extension direction with a maximum range of motion of 90 • /90 • (Flexion/Extension).

C. Tasks
The subject held the handle while their forearm was secured to the robot arm bracket with bandages (Fig. 2).Then the subject adjusted the sitting posture to make the elbow joint angle reach the most relaxed state (the range of elbow joint angle was 100±15 • ).The key concern was that the biological wrist position needed to be adjusted to match the wrist robot's joint rotation axis before the experiment started to avoid sampling errors caused by non-coincident joint rotation centers.
1) Workspace Determination: To determine if the wrist proprioceptive was related to the usage frequency of the joint, we divided the wrist ROM into different workspaces in an experimental way.Under the blindfolded condition, the subject holds the wrist robot handle for movement, and the wrist robot records the movement information.We defined the habitual workspace (HW) as the comfortable and effortless ROM commonly used by the subject during flexion/extension movement.The maximum workspace (MW) was defined as the subject moving the wrist as far as possible in the flexion/extension direction.In the experiment, subjects always started a movement task in a neutral wrist angle ("Origin" in Fig. 2).Subjects were first asked to complete 10 HW movement tasks, and 10 minutes later they were again asked to complete 10 MW movement tasks.The arithmetic mean of 10 tasks was considered the subject's workspace range.
2) Assessment: The ipsilateral joint position matching test was used to evaluate proprioception sensitivity.The robot drove the subject's wrist from the origin to the target position and returned after 3 seconds.Then, they were requested to move the wrist as much as possible to match the target position without assistance from the robot and repeated ten times.The arithmetic average of 10 times was taken as the test result.The vision of their wrist was occluded while the subject was actively operating the wrist robot.At this time, the motor provided an auxiliary force field to overcome friction in the gearing.It produced no torque or noise interfering with the subject's intention to move.The robot recorded the information on the wrist movement through the incremental encoder.The robot moves the wrist to the origin position before the start of each test.

3) Training:
The training procedure follows the assessment procedure but with audio feedback.During the training, the audio signals were used to indicate to the subject whether the flexion/extension of the wrist was beyond the target position, with a low pitch indicating that the wrist bending angle was lower than the target and a high pitch indicating that the wrist bending angle was higher than the target.The training process is shown in Fig. 2. The wrist was passively moved by the robot from the origin position to the target position, and the wrist was passively returned to the original position after 3 seconds.Then, the subjects actively manipulated the machine to match the target position.After 3 seconds, the high pitch prompt exceeds the target position.The prompt was given when the actual wrist position exceeded the target position.Finally, the wrist was brought back to its original position by the robot to start the next cycle of training.In addition, to prevent the adverse effects of fatigue on experimental results during the training process, we limited the maximum training duration to 30 minutes, and subjects were required to complete 45 position matching training tasks within 30 minutes.

D. Data Analysis
The wrist angle movement information before and after training was collected by the wrist robot.To better estimate the improvement of proprioception sensitivity before and after training, the difference between actual wrist motion position and expected wrist motion position was compared.The calculation of matching error (ME) can quantify the deviation of the actual wrist movement position from the target position.The calculation method is the average value of the error of repeated matching N (N=10) times for the same target position.
where σ ME is ME, θ i is the wrist's actual motion position, θ t is the expected position of the wrist, and N is the number of matching the target position.
The matching error reduction ratio (MErr) reflects the learning ability of proprioception.The larger the MErr, the more obvious the wrist proprioception is improved after training, and the stronger the learning ability of the wrist proprioception.
where σ MErr is MErr, σ MEa is the ME after training, σ MEb is the ME before training.All variables were presented with mean ± standard deviation and processed using SPSS22.0(Statistical Product and Service Solutions).Shapiro-Wilk (S-W) was used to test the normality of the before and after training data, using 2-way repeated measures ANOVA under the condition that the normal distribution was met, with p<0.05 being significant.

III. RESULTS
We collected the HW and MW of each person before the formal experiment, and the statistical results are shown in Fig. 3. Considering that the matching target position needs to be within the ability of each subject and cannot intersect between the two workspaces, we choose 80% of the workspace as the matching target position.These positions were: 38.51

A. The ME Before and After Training
The ME of the wrist to the target position of the 12 subjects before and after training were plotted as a radar Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.distribution map, As shown in Fig. 4.Moreover, as can be seen from Fig. 4, there was a clear difference between the flexion and the extension.The most wrists' ME to the target position were all reduced after training.Specifically, the ME of flexion to the target position was significantly reduced, and the ME of the same extension also has a downward trend.However, the decreasing trend was not evident compared with flexion after training.
The results of the S-W test showed no significance (p>0.05) for both that before and after training, which means that the collected data usually have distributed characteristics and the relationship between the differences before and after training can be analyzed using 2-way repeated measures ANOVA.The results of flexion in different workspaces are shown  II.After training, the wrist proprioception acuity was improved.The main effect of workspace was significant, F = 5.32, P = 0.031; the main effect of training was significant, F = 13.84,P = 0.001; the main effect of workspace * training was not significant F = 0.33, P = 0.57.Multiple comparisons revealed that the ME of before and after training in the HW was significantly lower, reaching a significance level (P = 0.006).In the MW, the ME of before and after training was significantly lower, reaching the significance level (P = 0.037).
As shown in TABLE III, the proprioceptive acuity of the wrist in different directions was improved after training in HW.The main effect of training direction was significant, F = 4.53, P = 0.045; the main effect of training was significant, F = 25.87,P <0.001; the main effect of direction * training was not significant F = 0.41, P = 0.53.The multiple comparison analysis results showed that the wrist's position matching error was significantly lower before and after training, with the flexion direction (P = 0.001) and the extension direction (P = 0.005) showing the significance.As shown in TABLE IV, the proprioceptive acuity of the wrist in different directions was improved after training in MW.The main effect of training direction was not significant, F = 3.82, P = 0.38; the main effect of training was significant, F = 15.22,P = 0.001; the main effect of direction * training was not significant F = 0.77, P = 0.39.The multiple comparison analysis results showed that the wrist's position matching error was significantly lower before and after training, with the flexion direction (P = 0.003) and the extension direction (P = 0.044) showing the significance.

B. Performance of Wrist Proprioception After Training
To further clarify the wrist learning ability, we analysed the MErr of wrist proprioception.As the results show in Fig. 5, wrist proprioceptive learning ability varied significantly for workspace and direction.In terms of the efficiency of wrist proprioceptive acuity enhancement, there was a significant difference(P <0.05) in the direction of wrist movement in both the HW(P = 0.037) and MW (P = 0.016).
The proprioceptive learning ability of the wrist flexion in different workspaces was different, and from Fig. 6a, the MErr in the wrist flexion was statistically significant (P=0.043) in the HW and MW.Fig. 6b showed that the proprioceptive learning ability of wrist extension was not significant in HW and MW(P = 0.268).

IV. DISCUSSION
"Brain plasticity" refers to establishing new connections between synapses and neurons due to learning and experience [20].Training can rebuild and strengthen nerve conduction in the brain, and improve proprioceptive acuity [21], [22].The improvement of proprioceptive acuity was reflected in the degree of joint-to-target position matching, and the smaller the ME of the joint-to-target position, the higher the proprioceptive joint acuity.The results shown in Fig. 4 showed a significant improvement in wrist joint matching to the target position after training, consistent with the results of previous research [23], [24].The results in TABLE I to TABLE IV show a significant difference between before and after training, confirming the beneficial effect of training on improving proprioceptive acuity.
In addition, it can be seen in Figures 4a and 4b that the proprioceptive acuity of the joint is improved more in HW than in MW.It can be seen in Figures 4a and 4c that the improvement in proprioceptive acuity is higher in flexion direction than extension for the same type of workspace.However, this trend is not evident in MW.Thus, we mapped the MErr of the wrist in each direction for different workspaces.The results are shown in Fig. 5.As shown in Fig. 5, the wrist proprioceptive learning ability was statistically significant in different workspaces and directions, which means there may be independent learning ability of the joint in different workspaces and directions.The proprioceptive acuity varies in different workspaces [14], [25].However, the proprioceptive showed different learning abilities in different workspaces under the same training conditions.Thus, we mapped the proprioceptive learning ability of the wrist in different workspaces and further analyzed the improvement of proprioceptive acuity of the wrist joint in different workspaces.From the results shown in Fig. 6a, the proprioceptive learning ability of 12 subjects Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
was significantly higher in HW than in MW, meaning that providing patients with a comfortable range of joint movement may accelerate the improvement of joint proprioceptive acuity.
Little research has been performed to clarify the relationship between the learning ability of joint proprioceptive and movement direction, and the previous experiments have only demonstrated the variability of proprioceptive acuity in different directions [13], [26].In this study, we found that the improvement efficiency of proprioceptive was related to the directions of the joint by analyzing MErr in each direction.The results in Fig. 5a and Fig. 5b show that the MErr was significantly higher in flexion than extension, which means that the joint has independent learning ability in different directions.Petrie's study showed that there are denser mechanoreceptors distributed in the radial collateral ligament of the wrist with a positional distribution that overlaps with the muscles of wrist flexion [27], and that these mechanoreceptors may provide a more sensitive spatial resolution for wrist flexion.However, the absolute error changes in different directions after training were insignificant, which may be because a slight variation in absolute error can lead to a large change in percentages.Therefore, prioritizing flexion training can improve the patient's proprioceptive acuity and help them build their confidence.At the same time, we should also pay attention to improving absolute errors in proprioceptive acuity after training and adjust the training strategy in time.
In addition, another of our findings may be helpful in restoring daily life to patients with functional disability.It is almost impossible for patients with functional disability to recover to their preimpairment level [28] fully, and the results in Fig. 3 may provide us with a criterion to demonstrate that the patient has restored the ability to live a normal life.We found that the ROM of the wrist in HW is close to 65% of the wrist's ROM in MW (33.58 • ≈31.29 • (65% of 48.14 • ); 48.14 • ≈51.17 • (65% of 78.72 • )).According to most people's habit of using wrist joints, 65% of ROM may be the key to helping patients return to normal life.
The paper proposed a study on wrist proprioception learning ability, using wrist robot-assisted technology to quantify wrist proprioception acuity accurately.The author also studied the relationship between training and proprioception acuity improvement.However, we are still unclear about the time and transfer effect that training can maintain proprioception acuity and the positive impact of different training methods on proprioception.We would venture to guess that the trend of the cumulative effect of training on proprioceptive acuity improvement may follow the memory forgetting pattern discovered by Hermann Ebbinghaus [29].These problems are worthy of attention in future research and need to accumulate more research to be resolved.In addition, the noise of the motor's continuous movement may help subjects complete the matching task.The subject determining the target position by the motor noise duration must fulfil at least two conditions.First, subjects could accurately memorize the duration of the movement, and second, the speed of the subject's wrist motion was consistent with the speed of the wrist robot's movement.However, the wrist robot will neither provide sound nor speed assistance in the target position matching task.Therefore, it is nearly impossible to complete the matching task by judging the duration of the movement.

V. CONCLUSION
This paper provides further insights into the increased proprioceptive acuity of the wrist in the workspace, which is very helpful for developing rehabilitation training strategies.The results of 12 healthy subjects showed that training while avoiding visual interference improves proprioception acuity.According to the performance of wrist proprioception after training, we found that the usage habit will accelerate the process of enhancing the proprioceptive acuity of the wrist, and the proprioceptive learning ability of flexion is better than that of extension.This finding is clinically beneficial to the formulation of patient treatment plans.Prioritizing the training of joints with strong learning ability helps patients build self-confidence.

Fig. 2 .
Fig. 2. The ipsilateral joint position matching test and training process.I passive motion, the wrist robot drives the wrist to the target position, and after 3 seconds it drives the wrist back to the original position.II active movement, the subjects actively matched the target position and received voice prompts.III passive movement, the wrist is brought back to the origin position.

Fig. 3 .
Fig. 3.The ROM of the wrist of 12 subjects.The sign of the ordinate indicates the direction, "-" indicates the extension; "0 • " indicates neutral wrist angle (original position); The blue line represents the average ROM of the wrist of 12 subjects (HW), and the yellow line represents the average ROM of the wrist of 12 subjects (MW).

Fig. 4 .
Fig. 4. The radar distribution map of the ME of the wrist to the target position before and after training.(a) Flexion in HW, (b) Flexion in MW, (c) Extension in HW, (d) Extension in MW.

TABLE I RESULTS
OF REPEATED MEASURES OF ANOVA FOR FLEXION IN DIFFERENT WORKSPACES TABLE II RESULTS OF REPEATED MEASURES OF ANOVA FOR EXTENSION IN DIFFERENT WORKSPACES in TABLE I.The results showed that the wrist proprioceptive acuity increased in different workspaces.The main effect of workspace was significant, F = 11.8,P= 0.002; the main effect of training was significant, F = 27.19,P<0.001; the main effect of workspace * training was insignificant F = 0.08, P = 0.78.Multiple comparisons revealed that the ME of before and after training in the HW was significantly lower, reaching a significance level(P = 0.001).In the MW, the ME of before and after training was significantly lower, reaching the significance level (P = 0.002).The results of extension in different workspaces are shown in TABLE

TABLE III RESULTS
OF REPEATED MEASURES OF ANOVA FOR THE WRIST IN DIFFERENT DIRECTIONS IN HW

TABLE IV RESULTS
OF REPEATED MEASURES OF ANOVA FOR THE WRIST IN DIFFERENT DIRECTIONS IN MW