From: Integration of proprioception in upper limb prostheses through non-invasive strategies: a review
Authors (Year) | Participants | Efference: End effector and control | Encoded info | Sensory Feedback: Stimulation—localization | Evaluation methods and results |
---|---|---|---|---|---|
Mann & Reimers (1970) [55] | 1 amputee | Myoelectric elbow prosthesis (Boston arm) | Elbow angle | Vibrotactile cutaneous display—Upper-arm socket | Improved accuracy (radial error) in matching tasks: comparable to standard mechanical prosthesis |
Hasson & Manczurowsky (2015) [54] | 27 able-bodied | Myoelectric virtual arm | Arm’s position and velocity | Vibrotactile (linear actuator)—Wrist of dominant arm | No improvements in end-point error, movement time and composite skill. In some cases, detrimental |
Guémann et al. (2022) [87] | 16 able-bodied and 7 amputees | Myoelectric virtual arm | Elbow angle | Vibrotactile array—1st third of upper arm/stump | Better performances in reach accuracy compared to no feedback. No further improvement with vision available but preferred according to NASA-TLX (workload evaluation questionnaire) |
Marinelli et al. (2023) [88] | 15 able-bodied and 1 congenital limb deficiency | Virtual wrist, manually controlled (keyboard keys) | Wrist angle (prono-supination) | Vibrotactile array (+ Gaussian interpolation spatial encoding)—Forearm | The novel method for continuous information encoding was demonstrated to be effective in providing the subjects with meaningful stimuli about prono-supination of the virtual wrist (average error < 10%). The approach described can be flexible customized in terms of sensation quality |
Vargas et al. (2021) [89] | 7 able-bodied | Myoelectrical finger | Finger MCP joint angle | Vibrotactile array—Upper arm | Subjects were able to control a myoelectric finger to the desired angle, both with a position- and velocity-based control, when provided with artificial proprioceptive feedback |
Witteveen et al. (2012) [90] | 15 able-bodied and 3 amputees | Virtual hand opening controlled by a scroll wheel | Hand opening (+ touch) | Vibrotactile vs Electrotactile—Forearm/Stump | In general, vibrotactile feedback performed better in grasping task considering duration, mean absolute error and % correct openings. Electrotactile feedback increases the duration. Addition of touch feedback further improves results |
Garenfeld et al. (2020) [86] | 13 able-bodied | Pattern classification-based Myoelectric control of virtual hand multiple DoFs | Wrist rotation and hand aperture | Electrotactile displays (spatial vs amplitude encoding)—Forearm (contralateral) | 2 different feedback configurations delivered good performances in terms of completion rate, time to the target, distance error and path efficiency |
Han et al. (2023) [91] | 5 able-bodied and 2 amputees | No control [Passive discrimination] | Flexion–Extension of prosthetic wrist | Electrotactile (static and dynamic coding for position and movement senses)—Forearm/Stump | Subjects were able to discriminate among 5 different degrees of flexion–extension of the prosthetic wrist, as well as the transition between one position and another (temporal combination of multiple position sensations). Lower success rates were reported for movement recognition |
Patel et al. (2016) [67] | 11 able-bodied | Myoelectric control of virtual and multi-articulated hand | Hand configuration | Electrotactile—Forearm (contralateral to control muscles) | Feedback improved performances in matching task, compared to no feedback condition. Results when visual feedback is available are always better |
Earley et al. (2021) [80] | 16 able-bodied | Hybrid positional-myoelectric control of a 2 DoF prosthesis | Joint speed | Frequency-modulated audio feedback | Joint speed feedback reduced reaching errors in ballistic reaching task, only when the control was perturbed by doubling the EMG gain |
Bark et al. (2008) [61] | 10 able-bodied | Hand-held force sensor-controlled cursor on screen (virtual arm) | Cursor position | Vibrotactile/Skin stretch—Forearm | Both feedback strategies improved position error results. Overall, skin stretch performed better |
Wheeler et al. (2010) [74] | 15 able-bodied | Myoelectric virtual arm | Elbow angle | Skin stretch (rotational)—Upper arm | Lower targeting errors with feedback, compared to no feedback. However, proprioceptive feedback from the contralateral limb obtained the best results |
Kayhan et al. (2018) [82] | 11 able-bodied | No control [Passive discrimination] | Wrist multiple DoFs-related information | Skin stretch—Forearm | Feasibility of the approach demonstrated by means of confusion matrices for discrimination accuracy |
Shehata et al. (2019) [99] | 1 amputee (lower limb) | No control [Passive discrimination] | Ankle movement | Skin stretch + TVI—Stump | Combination of feedback strategies increased the consistency of the illusion by 30–40%, depending on the site where the stretch was applied, with respect to TVI alone. Skin stretch also increases range and speed of the illusory movement |
Akhtar et al. (2014) [100] | 5 able-bodied | Myoelectric virtual fingers | Finger joint angle/hand configuration | Skin stretch/Vibrotactile—Forearm | Subjects were able to discriminate 6 different grips with 88% accuracy. In 1 DoF virtual targeting task, performances were similar between skin stretch and vibrotactile feedback |
Battaglia et al. (2017) [96] | 18 able-bodied | Myoelectric prosthetic hand | Hand aperture (prosthesis DC motor encoder) | Skin stretch (linear)—Upper arm | The feedback device was successfully integrated with the myoelectric prosthesis. Subjects’ discrimination accuracy improved with respect to no feedback condition |
Battaglia et al. (2019) [103] | 44 able-bodied and 1 amputee | Myoelectric prosthetic hand | Hand aperture (prosthesis DC motor encoder) | Skin stretch (linear)—Upper arm | Feedback was effective even in conditions of higher cognitive load (distraction task). Improved performances were reported in a functional test (AM-ULA) and passive discrimination test with the amputee user |
Rossi et al. (2019) [73] | 43 able-bodied and 1 amputee | Myoelectric prosthetic hand | Hand aperture | Skin stretch—Forearm | Improved discrimination accuracy both in passive (hand resting on table) and active (actively moving the hand) settings. 75% accuracy was reported, against 33% in no feedback condition |
Colella et al. (2019) [72] | 10 able-bodied and 1 agenesia | Myoelectric prosthetic hand | Hand aperture | Skin stretch (“unidirectional” and “pinch”—Forearm/Stump | Similar discrimination accuracies with different configurations of skin stretch were reported |
Pylatiuk et al. (2006) [106] | 5 amputees | Myoelectric prosthetic hand | Grip force | Vibrotactile—Stump | Improved ability in regulating the grasping force, with a reduction of 37–54% reported, compared to a vision-only condition |
Stepp & Matsuoka (2010) [107] | 8 able-bodied | Phantom Premium 1.0 robotic device (Sensable Technology) for 3D monitoring of the index fingertip | Applied Force | Vibrotactile/Haptic—Upper arm | Addition of vibrotactile feedback resulted in increased virtual box displacement and decreased difficulty ratings compared to vision-only condition, but performances were poorer compared to the ones obtained with the direct haptic feedback provided by the robot. Increased task times were reported with vibrotactile feedback |
Witteveen et al. (2014) [57] | 10 able-bodied and 7 amputees | Virtual hand controlled by computer mouse scroll wheel | Hand aperture and grip force | Vibrotactile (position + amplitude modulation)—Different configurations: Forearm and entire upper limb/ Stump | Subjects provided with opening information alone or in combination with force feedback were able to discriminate 4 stiffness levels in 60% of the cases. Performances were significantly better than those obtained without feedback. Force feedback alone was not sufficient to discriminate stiffness. No statistical differences between hand opening info alone and in combination with force |
Witteveen et al. (2015) [58] | 10 amputees | Virtual hand controlled by computer mouse scroll wheel | Hand aperture and grip force | Vibrotactile (position + amplitude modulation)—Stump | Performances were similar to those reported in the able-bodied group in a previous study |
Ninu et al. (2014) [105] | 9 able-bodied and 4 amputees | Myoelectric prosthetic hand | Hand closing velocity and grip force | Vibrotactile—Forearm/Stump | Vibrotactile feedback was effective in replacing visual feedback. Force feedback was not essential for the control of grip, given that subjects were able to do so predictively by means of closing velocity |
Jorgovanovic et al. (2014) [110] | 10 able-bodied | Virtual 1 DoF hand prosthesis controlled through a single-axis contactless joystick | Grasping force | Electrotactile—Forearm | The results showed that subjects were able to learn and scale the force based on electrotactile feedback, resulting in a 72% success rate in grasping objects of different weights. The closed-loop control demonstrated robustness and improved performance compared to feedforward control |
Chai et al. (2019) [104] | 15 able-bodied | Single DoF myoelectric hand prosthesis | Grasping angle and force | Electrotactile—Upper arm (ipsilateral to control muscles) | The feedback allowed subjects to discriminate among 4 types of grasped object sizes (87.5%), 3 kinds of stiffness (94%) and 4 levels of grasping forces (73.8%) |
Štrbac et al. (2016) [69] | 10 able-bodied and 6 amputees | No control [Passive discrimination] | Grasping angle and force, wrist rotation and flexion | Electrotactile (spatial + frequency coding)—Forearm/Stump | Subjects were able to discriminate stimulation levels with more than 90% success rate. Amputee subjects demonstrated lower rates than able-bodied |
Dosen et al. (2017) [64] | 10 + 10 able bodied | Myoelectric + Virtual hand prosthesis | Grasping force | Electrotactile (spatial/spatial + frequency coding)—Forearm (contralateral to control muscles) | The study suggests that mixed (spatial + frequency) coding is a reliable method for transmitting high-resolution information and offers advantages in terms of compactness compared to other coding schemes. Despite psychometric differences however, the performance in closed-loop control tasks was similar for both coding schemes |
Clemente et al. (2017) [81] | 8 able-bodied | Data glove-controlled robotic hand | Grip aperture and force | Augmented reality | AR feedback was successfully integrated into subjects’ sensorimotor control loop. Participants were also able to decouple the two types of information provided. AR feedback allowed the subjects to execute the task more consistently, compared to no feedback condition |
Dosen et al. (2015) [111] | 10 able-bodied and 2 amputees | Myoelectric prosthetic hand | EMG + Force/Force | Visual interface | The addition of EMG information reduced twofold force dispersion. More accurate and stable tracking of force was also reported. According to authors, force was controlled predictively (given the anticipatory nature of EMG signal) and with a finer resolution |
Schweisfurth et al. (2016) [68] | 11 able-bodied and 1 amputee | Myoelectric prosthetic hand | sEMG envelope vs Force (prosthesis input vs output) | Electrotactile (spatial + frequency coding)—Forearm (contralateral to control muscles) | EMG allows for predictive control (improved feedforward control) and improved precision of myoelectric command and force control |