Low cost exoskeleton manipulator using bidirectional triboelectric sensors enhanced multiple degree of freedom sensory system

Rapid developments of robotics and virtual reality technology are raising the requirements of more advanced human-machine interfaces for achieving efficient parallel control. Exoskeleton as an assistive wearable device, usually requires a huge cost and complex data processing to track the multi-dimensional human motions. Alternatively, we propose a triboelectric bi-directional sensor as a universal and cost-effective solution to a customized exoskeleton for monitoring all of the movable joints of the human upper limbs with low power consumption. The corresponding movements, including two DOF rotations of the shoulder, twisting of the wrist, and the bending motions, are detected and utilized for controlling the virtual character and the robotic arm in real-time. Owing to the structural consistency between the exoskeleton and the human body, further kinetic analysis offers additional physical parameters without introducing other types of sensors. This exoskeleton sensory system shows a great potential of being an economic and advanced human-machine interface for supporting the manipulation in both real and virtual worlds, including robotic automation, healthcare, and training applications.

. Photos of the as-fabricated sensors and the assembled exoskeleton arms. a (i) Main components of triboelectric bi-directional (TBD) sensor, (ii) assembled TBD sensors, including two different sizes, and (iii) assembled linear TBD sensor. b (i) Integration of the sensors on the exoskeleton arm and glove, and (ii) major components of the exoskeleton arm, including forearm, upper arm, L-shaped shoulder module, back support (from left to right). c Exoskeleton after wearing (i) front view, (ii) back view, and (iii) side view and schematics of exoskeleton with the rotational triboelectric bi-directional back (RTBD-B) sensor, the rotational triboelectric bi-directional shoulder (RTBD-S) sensor, the rotational triboelectric bi-directional elbow (RTBD-E) sensor, the rotational triboelectric bi-directional wrist (RTBD-W) sensor, and the linear triboelectric bi-directional finger (LTBD-F) sensor. Photo credit: Minglu Zhu, National University of Singapore.

Supplementary Note 1. Optimization strategy for high resolution and high rotation speed sensing
As mention in the main manuscript, the small spacing of the grating pattern experiences the severe loss of output peaks at high rotation speed, i.e., 300 revolutions per minute (RPM), as shown in Supplementary Fig. 3a. Compare to the signals at 200 RPM, this issue of the resolution loss will affect the signal processing for manipulation purpose, and cause the inconsistency between the actual human motion and the controlled objects. The main reason of this problem is the width of the copper spring which is used to slide across the PTFE gratings ( Supplementary Fig. 3b(i)). As the spacing keep decreasing, there will be no enough space for the copper spring to achieve a full separation with the previous grating, as illustrated in Supplementary Fig. 3b(ii). In another word, the copper spring will connect both of the adjacent gratings, which will affect the charge transfer due to the incomplete contact-separation cycle. Hence, to solve this issue, an effective approach is to reduce the width of the copper spring, in order to re-establish the complete contact-separation cycles. (i) 20℃ and 95% relative humidity, (ii) 20℃ and 50 % relative humidity, (iii) 48℃ and 20% relative humidity. Rotation speed: 100 revolutions per minute (RPM).

Supplementary Note 2. Response latency for the switching of the direction
As a bi-directional sensor with a pendulum-like switching mechanism, the design of the switch structure may affect the response latency of the sensor, and eventually determine the state of synchronization between the human motion and the controlling objects. To investigate this issue, two designs of switches were proposed. The switch with shaper corner has the gaps of 1.5 mm at neutral state, where as the switch with flatter angle has 3 mm gaps. In order to generate the triboelectric output for monitoring the switching process, the PTFE as negative triboelectric materials were attached on E1 and E2, and the aluminum film on switch was positive triboelectric materials. Both of two switches were tested under two rotation speed, 10 RPM and 100 RPM. As shown in Supplementary Fig. 6, the red negative peak represents the separation of E1, and the black positive peak indicates the contact of E2. Based on the enlarged view of the peaks of the separation and the contact during switching, the response latency is about a few tens of milliseconds, which can be considered as small for the practical applications. These gaps can also be further reduced by modifying the design of the switch. Supplementary Fig. 6. Evaluation of response latency during the switching of the direction for different designs of switches. a Switch with 1.5 mm gap at neutral state, and the corresponding signals from E1 and E2 electrodes during the switching, the red negative peak represents the separation of E1, and the black positive peak indicates the contact of E2, the enlarged graph of the dashed box, as well as the time and the peak voltages of the separation (1) and contact (2) signals are provided. b Switch with 1.5 mm gap at neutral state, and the corresponding signals from E1 and E2 electrodes during the switching, the enlarged graph of the dashed box as well as the time and the peak voltages of the separation (3) and contact (4) signals are provided.

Supplementary Note 3. Customized print circuit board (PCB) for optimizing multichannel sensing of triboelectric output
The direct triboelectric signal readout from the commercialized microprocessor with a single analogto-digital converter (ADC), such as Arduino, will engage with the cross-talk and the large noise issues. As shown in Supplementary Fig. 7a, the conditioner unit consists of operational amplifier and low pass filter is applied to solve those problems, while maintaining the floating signal to acquire the entire triboelectric waveform in Supplementary Fig. 7b(i). Furthermore, the unified output peaks are preferred to ensure the convenient programming of the peak detection. Hence, the comparator circuit is also introduced to convert the original pulse waveform into the square waveform ( Supplementary  Fig. 7b(ii)). Next, by utilizing the edge inspection code to detect those square peaks, the rotation signals can be expressed into the desired command, i.e., 1 for one peak from clockwise rotation, 2 for one peak from counter-clockwise rotation. With the TCP/IP serial communication between Python and Unity, the virtual character can then be controlled in real-time. For robotic manipulation, the command with angular information need to be converted into hexadecimal format before send to the servo controller. A communication cable for USB to TTL is used for data transmission. Supplementary Fig. 7. Signal processing circuit and transmission for the multichannel sensors. a Flow chart of signal processing and transmission. b (i) triboelectric signals before the comparator, and (ii) after comparator.

Supplementary Note 4. Kinetic analysis of punching force estimation via rotational triboelectric bi-directional (RTBD) sensors
The straight punching is selected for a simplified analysis of the boxing dynamics. In this case, the shoulder joint is considered as a fixed point by assuming no torso motion. The path of punching is a direct forward punching, and hence, the factor of eccentricity c equals to 0 since all of center of three joints (shoulder, elbow, and fist) are aligned in a horizontal line. In general, this model can then be treated as a slider crank mechanism for further investigation. As illustrated in Supplementary Fig. 8a and 8b, the length between the shoulder and the fist can be expressed as: = 1 cos + 2 cos (1) Whereas L is the length between the shoulder and the fist, L1 is the length of the upper arm, L2 is the length of the forearm, α is the angle between the upper arm and the center line of punching, β is the angle between forearm and the center line of punching.
As the largest distance the shoulder and the fist is: = 1 + 2 (2) Hence, the displacement of the fist can be calculated as: = = ( 1 cos + 2 cos ) (3) Let = 1 / 2 = 1 cos + 2 (√1 − (sin ) 2 2 ) By utilizing the function of rotation speed sensing of the proposed sensors, the angular speed can be detected as , , for the rotation of shoulder and elbow, respectively. Meanwhile, the velocity of the fist can be determined as: Since = , the velocity of the fist can then be calculated as: Furthermore, the acceleration of the fist can be calculated as: Next, based on conservation of momentum, two equations (Eq. 4 and Eq. 5) are applied to define the forces exerted from the fist and received by the target, respectively. = (8) − 0 = (9) Where m is the mass of the whole arm, a is the acceleration of the fist, v is the final velocity when contacting with target, v0 is the initial velocity at the beginning, Ff is the forces generated from the fist, Ft is the force received by the target, t is the time period from the contacting of the target to the complete stop. To verify the proposed kinetic analysis for the estimation of the punching force, a force meter is applied to determine the errors. As shown in Supplementary Fig. 8a(iii), the punching forces of 4 kg and 6 kg were measured for the linear fist velocity of 1 m/s and 1.5 m/s, respectively. By assuming the contact time period as 0.1 s, the calculated force are 40 N and 90 N, respectively. According to Fig. 6g(ii), the estimated forces (50 N and 87 N) via the data of the rotation TBD sensor can be considered as reasonable, with the errors of 25% and 3%.