The LIBRA NeuroLimb: Hybrid Real-Time Control and Mechatronic Design for Affordable Prosthetics in Developing Regions

Abstract

Globally, 2.5% of upper limb amputations are transhumeral, and both mechanical and electronic prosthetics are being developed for individuals with this condition. Mechanics often require compensatory movements that can lead to awkward gestures. Electronic types are mainly controlled by superficial electromyography (sEMG). However, in proximal amputations, the residual limb is utilized less frequently in daily activities. Muscle shortening increases with time and results in weakened sEMG readings. Therefore, sEMG-controlled models exhibit a low success rate in executing gestures. The LIBRA NeuroLimb prosthesis is introduced to address this problem. It features three active and four passive degrees of freedom (DOF), offers up to 8 h of operation, and employs a hybrid control system that combines sEMG and electroencephalography (EEG) signal classification. The sEMG and EEG classification models achieve up to 99% and 76% accuracy, respectively, enabling precise real-time control. The prosthesis can perform a grip within as little as 0.3 s, exerting up to 21.26 N of pinch force. Training and validation sessions were conducted with two volunteers. Assessed with the "AM-ULA" test, scores of 222 and 144 demonstrated the prosthesis's potential to improve the user's ability to perform daily activities. Future work will prioritize enhancing the mechanical strength, increasing active DOF, and refining real-world usability.

Keywords: brain–computer interface; machine learning; myoelectric control; pattern recognition; sensor fusion; transhumeral prosthesis.

Figure 1

Internal and external views of the prosthesis: (a) hand prosthesis assembled, (b) internal components.

Figure 2

Grip types and gestures: (a) cylindrical, (b) hook, (c) pinch, (d) finger-pointing/pressing.

Figure 3

Finger mechanism: (a) previous underactuated mechanism with joints “A”, “B”, “C”, and the top of the fingertip “D” (b) finger transition, (c) points “A”, “B”, “C”, and “D” trajectory in a sagittal plane.

Figure 4

Index finger and finger group gear transmission: (a) servo and index finger connection, (b) index finger maximum extension, (c) index finger maximum flexion, (d) servo and fingers connection, (e) fingers maximum extension, (f) fingers maximum flexion.

Figure 5

Thumb mobility and parts: (a) thumb flexion/extension, (b) thumb abduction/adduction.

Figure 6

LED indicators, sEMG cables, and power connector.

Figure 7

Forearm: (a) components, (b) designs.

Figure 8

Wrist connected to the hand: (a) adduction/abduction trajectory, (b) flexion/extension trajectory considering 90-degree wrist connection alternation.

Figure 9

Wrist principal parts: (a) wrist base and rotation positions, (b) wrist and the principal buttons.

Figure 10

Forearm upper detail: (a) internal battery and its supports, (b) additional components from the battery cover, (c) elbow locking system, (d) battery cover locking system.

Figure 11

Elbow and arm: (a) elbow locking system, (b) arm tilt positions.

Figure 12

Body attachments designed: (a) transhumeral socket, (b) short transhumeral socket.

Figure 13

Main electrical components: (a) internal palm components, (b) external BCI device.

Figure 14

System architecture. Abbreviations: analog-to-digital converter (ADC), pulse width modulation (PWM), electroencephalography (EEG), and surface electromyography (sEMG).

Figure 15

Simplified flow diagram of the hybrid control system in the LIBRA NeuroLimb. Abbreviations: brain–computer interface (BCI), electroencephalography (EEG), surface electromyography (sEMG), and support vector machine (SVM).

Figure 16

Cubic SVM 5-fold cross-validation accuracy heatmaps based on C and Gamma hyperparameters: (a) LV, (b) OB, and (c) LV and OB combined data.

Figure 17

k-nearest neighbors 5-fold cross-validation accuracy heatmaps based on the number of neighbors, weight function (U for Uniform, D for Distance), and distance metric (E for Euler, M for Manhattan): (a) LV, (b) OB, and (c) LV and OB combined data.

Figure 18

Neural network 5-fold cross-validation accuracy heatmaps with respect to single layer units and Adam optimizer learning rate: (a) LV, (b) OB, (c) LV and OB combined data.

Figure 19

SVM training and validation accuracy curves for training examples with 5 features: (a) LV’s data, (b) OB’s data.

Figure 20

Training and validation metrics of a wide neural network trained with LV, OB, and combined datasets: (a) loss curves, (b) accuracy curves. Parameters: learning rate = 0.1, units = 100, epochs = 100, hidden units activation = ‘relu’, output activation = ‘softmax’, loss = ‘categorical crossentropy’.

Figure 21

Prosthesis gestures: (a) cylindrical gesture, (b) hook gesture, (c) cylindrical gesture, (d) finger-pointing gesture, (e) cylindrical grip, (f) hook grip, (g) cylindrical grip, (h) finger-pressing gesture.

Figure 22

EEG classification algorithm accuracy across participants (LV and OB) and commands (ON and OFF) per repetition. P001 and P002 refer to LV and OB, respectively.

Figure 23

Volunteer OB performing training and AM-ULA sub-tasks: (a) Training with EEG sensors. (b) Simulated drinking. (c) Holding his cell phone. (d) Zipping sub-task. (e) Holding a cup of water. (f) Reaching an object on a shell.

Figure 24

Volunteer LV performing training and AM-ULA sub-tasks: (a) sEMG sensors placement measure on amputated section. (b) Training with EEG sensors. (c) holding a cup of water. (d) Using cutlery. (e) Writing initials on a board. (f) Folding a piece of cloth.