*3.3. Testing*

The whole mechatronic system described in the previous sections has then been assembled and tested to evaluate the manufacturability of the adopted solution (see Figure 2). The single DOF mechanism, the one visible in Figure 1, consists in a closed kinematic chain mechanism capable of applying a defined roto-translation motion to the middle point of the intermediate phalanx. Again, Kinovea software has been employed in this testing phase. The exoskeleton has been worn by the user who has been told to perform some opening and closing gestures. Colored markers have been placed on the mechanism joints and their trajectories have been extracted from the images. Finally, the acquired trajectories has been compared to the kinematic model ones. The comparison between real data and simulated trajectories are presented in Figure 4 only for the index finger mechanism. The same considerations can be adopted for all other long fingers.

**Figure 4.** The figure shows the comparison between the trajectories of the joints of the exoskeleton generated by the kinematic model (solid lines) and the real ones (dashed lines) for the index finger.

Despite the fact that the comparison of the trajectories has given encouraging results—in particular with regard to the joint 2, which is the main point of interaction between the exoskeleton and the finger—tests conducted on this first version highlighted several flaws which negatively impacted on its usability.


• The solution provided by the exploited optimization algorithm was, by the nature of the algorithm itself, strongly dependable on the choice of the initial state, which was arbitrarily made accordingly to the imposed geometric constraints. This process hence resulted in being low-adaptable to different hand sizes because of the necessity of proper setting the initial state of the algorithm every time, and it also offered no guarantees of finding a real optimum solution.

The tested device hence resulted in being still far from a clinical application and many improvements had to be made to move towards that direction.

#### **4. Second Prototype: Ergonomics and Adaptability Improvements**

Starting from the weaknesses of the exoskeleton version discussed in Section 3 and considering that the adaptability and the ease of wearing requirements must be, not only kept, but also improved, a second prototype of the hand exoskeleton has been developed and printed. In particular, thanks to the experience gained during the testing phase several modifications have been made to produce a more comfortable and wearable system, and develop a new design procedure easier to adapt to different users. The high wearability of a system is hence endorsed by its transparency with respect to the hand natural kinematics, rising from a device which is not felt as a constraint, but whose use actually results straightforwardly intuitive. Both the optimization strategy, leading to a mechanism which replicates at best the fingers gestures, and the revamping design, heading to a lighter solution, contribute to a system better accepted by the end user.

## *4.1. Mechanical Design*

The optimization algorithm, in charge of modifying the geometry of the device accordingly to the fingers trajectories has been replaced: a Nelder-Mead based optimization algorithm, used to solve non convex, non linear constrained problems, has been applied achieving a straightforward adaptability to several users [37]. Taking acquisition data (collected exploiting a BTS SMART-Suite MoCap System by BTS Bioengineering (BTS Bioengineering, Garbagnate Milanese (MI), Italy)) and the kinematics of the mechanism as inputs, the implemented algorithm provides a customized geometry specific for each patient.

A CAD model has then been developed by using parametric dimensions to make the whole procedure automatic. This choice leads to two important consequences. Firstly, the position of each joint results directly connected to the outputs of the optimization routine without requiring manual intervention. In addition, as well as tested with the first prototype, a parametric CAD model enable wearing simulations. By replicating virtual opening and closing gestures with the device worn by the user, the exoskeleton kinematics and its coupling with each finger can be checked and assessed. The ABS prototype is then print out exploiting the Fused Deposition Modeling (FDM) additive manufacturing technique.

Even though the overall kinematics of the finger mechanism has not been changed with respect to the first prototype, the mechanical architecture of the exoskeleton has been subjected to important modifications. The thimble wrapping the finger tip has been eliminated leaving only one contact point between each finger and the device. By removing this component, the driving purpose of the finger mechanism has been kept but the uncomfortable feeling of having the whole finger constraint to a rigid system has been avoid. This choice also leaves the touching feeling while grabbing objects. A passive DOF has been added upstream joint 1 (refer to Figure 1) to follow finger abduction and adduction motions during hand opening and closing. This solution also improves the auto-alignment between finger and mechanism joints.

#### *4.2. Actuation System and Control Strategy*

To ge<sup>t</sup> closer to the needs come out during the tests on the previous prototype, this new "release" of the hand exoskeleton presents only two servomotors Hitec HS-5495BH (http://hitecrcd.com/products/servos/sport-servos/digital-sport-servos/hs-5495bh-hv-digitalkarbonite-gear-sport-servo/product): one of them guides the index finger flexion/extension, the other one is in charge of moving the other three long fingers. In addition to reducing the overall weight and encumbrance of the system, reducing the number of actuators overcome an issue that has immediately occurred the high difficulty to prevent the four motors from acquiring more and more relative phase shift with the prolonged use.

Two motors, one the other side, require an important change in the transmission system: because of the different sizes of each finger mechanisms (due to, in turn, the dissimilar dimensions of the fingers), closing and opening velocity for each finger must be different one to the others. To overcome this issue, the pulley spliced to the output shaft of the actuator in charge of moving three long fingers has been thought with three different diameters. That introduced a set gear ratio and, since the wires connected to the three fingers mechanisms winds pulley with different diameters, middle, ring and small finger mechanisms are moved at the same time with the same motor but each one at its own suitable speed. The overall weight of this new prototype resulted in 242 g. Regarding the characteristics of the new actuators, they present a maximum torque of 0.735 Nm at 7.4 V with a size of 39.8 × 19.8 × 38.0 mm and a weight of 44.5 g. The maximum angular speed is 66.67 rad/s at 7.4 V.

Also the control system has been developed aiming to the same goal headed to during the mechanical design: the total costs, complexity and weight of the system must be kept as low as possible. In this framework, Arduino Nano represented a simple, cheap, but performing solution. Two magnetic encoders (15-bit resolution) are spliced on joint 1 of the finger mechanisms (Figure 5). Since one servomotor actuates the index and the other one the other three long fingers, in order to guarantee a suitable control of the grasping, only two encoders are needed: one placed on the index and one on the little finger. They measure the value of the angle *α*2, which identifies pose of the mechanism and, consequently, of the finger fixed to it.

**Figure 5.** The figure shows, on the left, the second version of the exoskeleton system worn by a healthy subject and, on the right, the corresponding kinematic chain and CAD model. Colors and names (capital letters) of the components, and joints enumeration are reported as introduced in Section 2.

The control strategy is characterized by two main control loops that have been added to safely stop the motion of the exoskeleton once the ROM limits are reached (namely maximum opening and closure), and to check if an object is grabbed. While the permanence within the ROM boundaries is easily checked by means of the direct measurement from the encoders, the grabbing of an object is recognized and detected by the evaluation of the length of the unrolled cable twice. The first evaluation is made by the kinematic equations of the mechanism and the second one exploiting the motors speed and the pulleys radius. Comparing the differential measurement to a fixed threshold (set at 10 mm, which is roughly the length of a quarter of the motor pulley circumference) allows to understand if the actuation system is still releasing cable while the hand is not further moving. This means that an object is likely grasped. In both cases each motor stops while holding its current angular position.
