Haptic Knee Brace

Johns Hopkins University · 535.691 Haptic Interface Design · Prof. Jeremy Brown

Wearable haptic brace providing directional vibrotactile feedback for ACL rehabilitation, guiding users toward stable knee positions without constraining movement.

Role: Hardware lead
Team: Team of 4
Tools: Arduino Mega / Nano · PWM motor drivers · Flex sensors · Fusion 360 · 3D Printing (TPU, PLA)

What I did

Designed the electronics: Arduino Mega + Nano link, transistor drivers for 6 vibration motors and a connectorised harness
Integrated 3 flex sensors with calibration for 5°/8° bend detection, in 3D-printed TPU and PLA housings
Defined the directional vibration logic the team coded, and co-ran user studies

KICD wearable haptic knee brace with flex sensor cage and ERM motor array worn on leg — KICD device worn on leg showing flex sensor cage around knee joint and ERM motor band on upper thigh

Problem Statement

Knee instability from ACL injuries affects 80,000–120,000 people annually in the US. Current rehabilitation methods face a tradeoff.

Over-bracing risk: Passive braces limit movement, which leads to muscle atrophy
Under-bracing risk: Free movement during recovery risks re-injury
Dynamic valgus: Lateral knee collapse during movement is primary injury mechanism
Gap: No device provides active guidance without physical constraint

Methodology

Complete wearable system with sensing, actuation, and closed-loop control.

Design & Fabrication

The device consists of two adjustable bands: a lower band with flex sensors around the knee and an upper band with the ERM motor array on the thigh.

Flex sensors: 3D-printed TPU enclosures for durability and compliance
ERM motors: 1020-type, 79–183 Hz frequency range, 40mm spacing for ≥75% discrimination
Bands: Adjustable elastic with Velcro attachment

KICD mechanical design showing flex sensor enclosures and ERM motor mounting — Flex sensor enclosures (TPU) and ERM motor array mounting configuration

Electronics Architecture

The system uses an Arduino Mega as the main controller, commanding all ERM motors and receiving data from ELEGOO Nano boards that interface with each flex sensor.

Main controller: Arduino Mega for motor control and data aggregation
Sensor interface: ELEGOO Nano boards for each flex sensor
Motor drivers: 2N2221A NPN transistors, PWM-controlled
Wiring: Two perfboards with connectorized bundles for serviceability

Electronics architecture: Arduino Mega controller with ELEGOO Nano sensor interfaces and transistor motor drivers — Electronics architecture showing Arduino Mega, Nano sensor interfaces, and transistor motor drivers

Control Logic

Threshold-based detection with staggered "follow me" vibration patterns that intuitively guide users toward stability.

Detection: Flex sensors detect knee angle across sagittal, abduction, and adduction planes
Thresholds: 5° triggers warning, 8° triggers active correction feedback
Feedback: Staggered "tracing" vibration pattern guides user directionally
Calibration: Auto-calibration routine on startup for per-user adaptation

Control logic diagram showing threshold-based detection — Threshold detection logic

Staggered vibration pattern visualization — Staggered vibration pattern

Test Setup

Preliminary vibrotactile perception studies determined optimal design parameters.

4 Subjects (discrimination)

9 Subjects (intensity)

30+ Demo subjects

22±11° Peak abduction angle

Two-point discrimination: Testing across 4 knee regions established 40mm motor spacing
Intensity perception: Characterized nonlinear PWM-to-sensation relationship
Device validation: Tested on 2 individuals with uninjured knees
Haptic demos: 30+ subjects using haptic twin configuration for safe manipulation

Results

Intensity perception curve showing 2nd order polynomial relationship — Intensity perception curve (R²=0.999) characterizing nonlinear PWM-to-sensation relationship

            Data
            Key findings from testing
          
            100% accuracy: Front knee bend detection and haptic triggering
50% accuracy: Compounded abduction direction (requires per-user calibration)
≥75% discrimination: Achieved above 24mm spacing, near-perfect above 40mm
R²=0.999: Intensity perception follows 2nd-order polynomial curve
6% knismesis: Subjects experienced tickling sensation preventing response
Clothing: Vibration perceivable through clothing with no significant damping

          

Table Two-point discrimination by region

Region	Location	24mm Accuracy	40mm Accuracy
1U	Front	≥75%	~100%
2U	Outer	≥75%	~100%
3U	Back	≥75%	Variable
4U	Inner	≥75%	~100%

Future Planned improvements

Voice coils: Replace ERMs for finer frequency control
2-axis flex sensors: Better abduction/adduction detection
Accordion TPU mounts: Improved mechanical compliance
Formal user study: IRB-approved clinical validation

Demo Video

Demonstration of the KICD device mounted on a teddy bear mannequin, showing flex sensor detection of incorrect knee positions and corresponding directional haptic feedback response.

Wearing It

Wearing it was interesting, and you could clearly feel the vibration cues. What we couldn't do was feel it working as intended, because none of us have knees that bend in the direction the device is meant to correct.

Full Report

Complete technical report with methodology, results, and analysis from the 535.691 Haptic Interface Design course final project.

Skills Technical skills demonstrated

Electronics: Arduino, transistor motor drivers, PWM control
Fabrication: 3D printing (TPU), wearable integration
Haptics: Vibrotactile perception, two-point discrimination
User Studies: Perception testing, data analysis