ElectroGrasp: Electrotactile Aids for Visually Impaired Individuals in Anticipatory Planning and Control of Grasp
Hechuan Zhang, Rufei Song, Ruoyan Liu, Shengsheng Jiang, Xiaohui Tan, Tianren Luo, Yulin Jin, Hongnan Lin, Teng Han, Feng Tian · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790930
Summary
Sighted people pre-shape their hand before touching an object - a visually-driven anticipatory grasp plan that visually impaired (VI) users cannot form, leading to the characteristic 'double grip' (initial contact followed by compensatory regrasp) that is slower, less confident, and sometimes knocks objects over. ElectroGrasp is a wearable electrotactile prototype that delivers anticipatory spatial information about an upcoming object directly to the skin of the hand, encoding three properties before contact: grasp orientation (horizontal vs vertical, conveyed by four tested stimulation patterns across fingertips and palm), object size (finger-flexion target conveyed by frequency-modulated stimulation that tapers as the hand closes), and object shape (contact-region patterns derived via ResNet-18 CNN features and clustering on the STAG pressure-map dataset into ten representative shape classes). The hardware is ambitious: 208 stimulation sites across flexible printed circuit arrays on all five fingers plus two palm regions, driven by a custom high-voltage (up to 155V, 5mA safety limit) multiplexed driver controlled from a PC tracking finger flexion via a Leap Motion. The authors evaluate ElectroGrasp across three experiments with up to 17 VI participants (mix of congenitally blind, acquired blind, and low-vision): Experiment 1 calibrates pattern discriminability and size just-noticeable-difference thresholds; Experiment 2 measures learning cost for object-pattern associations; Experiment 3 compares grasp kinematics under audio cues versus electrotactile cues for ten real objects.
Key findings
Experiment 1: participants reliably discriminated all 10 shape patterns within about 8 seconds (accuracy 0.82-1.0 per pattern); the finger-flash mode for orientation encoding achieved 100% accuracy and was preferred by 75% of participants over sweep-based alternatives. Size just-noticeable-difference was 0.67 for large-flex and 0.33 for small-flex (no significant difference between congenitally blind, acquired blind, and low-vision groups on size sensitivity). Experiment 2: a treatment group given 10.5 minutes of training reached 87% object-recognition accuracy versus 53% for an untrained control group (p = 0.001), and recognition times dropped substantially - a learning cost the authors argue is comparable to standard haptic-device familiarisation. Experiment 3: electrotactile cueing eliminated overshoot events (0 vs 0.33 average under audio) and significantly reduced regrasps from 1.93 to 1.61 (p = 0.014), with grasp execution time dropping from 3.28s to 2.96s. NASA-TLX showed electrotactile cueing imposed higher physical, mental, and temporal demand than audio (attributed to novelty), but no difference in performance or frustration. Strikingly, congenitally blind participants described the stimulation as 'a direct replacement for vision' and found it more intuitive than low-vision participants, who preferred residual vision and treated tactile cues as supplementary - suggesting electrotactile substitution benefits scale with visual deprivation.
Relevance
For accessibility practitioners this paper is interesting not for the hardware (which is clearly lab-bound - 208 sites, wired drive unit, external Leap Motion) but for what it establishes conceptually: cues delivered during the pre-grasp phase can materially change motor strategy, shifting VI users from a reactive grab-then-adjust pattern to planned pre-shaping that looks kinematically like sighted behaviour. That reframes what assistive haptics should aim for - most existing wearables answer 'where is the object?' (navigation) but few answer 'how should I hold it?'. The differential response by visual-experience group is the more practically important finding: a system optimised for fully blind users may feel redundant or intrusive to low-vision users, and vice versa, so adaptive or personalised cueing is essential. Limitations are significant: the device is wired to external hardware, participants were young (mean age 26) university-recruited, training was brief, and the study sites are all in controlled lab conditions. Real-world deployment - on unfamiliar objects, in public spaces, over sustained use - remains an open question, as does self-contained wearable form factor (the authors flag embedded IMUs/bend sensors as the path forward).
Tags: blind and low vision · electrotactile feedback · haptic technology · assistive technology · wearable technology · spatial perception · grasping · proprioception · sensory substitution · motor control · cross-modal plasticity