Auricular Muscle- controlled Navigation for Powered Wheelchairs

This paper introduces and experimentally validates a novel, socially unobtrusive navigation system for powered wheelchairs that utilizes vestigial auricular muscle EMG signals, demonstrating through a comparative study of control strategies and a three-participant trial that a 300ms windowed Support Vector Machine approach offers a viable and accurate real-time control method for users with tetraplegia.

The Gist

Imagine you are trying to drive a car, but your hands and feet are tied up. You need a way to steer, but the usual methods—like blinking, puffing air through a straw, or moving your head—feel awkward, slow, or make you feel self-conscious in front of other people.

This paper is about finding a "secret superpower" hidden in your ears to drive a wheelchair.

The Problem: The "Socially Awkward" Controls

Currently, many people with severe paralysis (like tetraplegia) use wheelchairs controlled by joysticks, head movements, or even blinking. While these work, they have a big downside: they break the flow of conversation.

Imagine trying to have a deep conversation with a friend, but every time you want to turn a corner, you have to stop talking, puff your cheeks, or blink rapidly. It's distracting and makes you feel like a robot rather than a person. The researchers wanted a control method that is invisible to others, so the user can focus on being a person, not just operating a machine.

The Solution: The "Ear Wiggle" Muscle

Enter the Auricular Muscles. These are the tiny muscles around your ear that allow some people to wiggle their ears. For most humans, these are "vestigial" muscles—like your wisdom teeth or tailbone. They are leftovers from our evolutionary past (our ancestors used them to swivel their ears like dogs to hear predators) and we generally don't use them anymore.

The Big Idea: Because these muscles are controlled by a different nerve than your face, even people with severe spinal cord injuries who can't move their face might still be able to wiggle their ears. Plus, since they are hidden inside the ear, wearing a sensor there is like wearing a secret earpiece—it doesn't look like a medical device.

The Experiment: Two Ways to Drive

The researchers built a small robot wheelchair and taught three healthy volunteers (who could wiggle their ears) how to drive it using two different "languages" of ear wiggles.

1. The "Gas Pedal" Method (Continuous Control Strategy)

Think of this like driving a car with your foot on the gas.

• How it works: If you wiggle your left ear, the robot turns left. If you wiggle your right ear, it turns right. If you wiggle both, it goes forward. If you stop, it stops.
• The Analogy: It's like holding a steering wheel. You have to keep your muscles engaged to keep moving.
• The Result: This felt very intuitive and natural to the users. However, it required a bit more effort to keep the muscles "on" the whole time, which made some users feel a bit more tired.

2. The "Morse Code" Method (AM-MCWN)

Think of this like sending a text message with short and long beeps.

• How it works: You use just one ear.
  • A short wiggle means "Turn Left."
  • A long wiggle means "Turn Right."
  • Two short wiggles = Go Forward.
  • One short, one long = Go Backward.
• The Analogy: It's like tapping out a secret code on a table. You tap, wait for the robot to understand, then tap again.
• The Result: This was less tiring because you only wiggle briefly. However, users found it harder to learn because they had to remember the code and wait for the robot to process it. It felt a bit more like playing a video game than driving a car.

The "Brain" Behind the Ear

The robot doesn't just "hear" the wiggle; it has a brain (a computer algorithm called a Support Vector Machine) that listens to the electrical signals from the ear muscles.

The researchers had to figure out the perfect "listening window."

• If the computer listens for too long (like waiting 2 seconds for a wiggle), the robot feels sluggish and slow.
• If it listens for too short a time (like 30 milliseconds), it might mistake a random twitch for a command.
• The Sweet Spot: They found that 300 milliseconds (about the time it takes to snap your fingers) was the perfect balance. It's fast enough to feel responsive but slow enough to be accurate.

What Did They Learn?

1. It Works: The system successfully navigated a maze.
2. It's Personal: Different people liked different methods.
   • If you are good at multitasking and have strong muscle control, the "Gas Pedal" (Continuous) method is faster and more intuitive.
   • If you get tired easily or only have control over one side of your body, the "Morse Code" method is better because it requires less effort.
3. The Hiccup: The biggest problem wasn't the brain or the muscles; it was the hair. The sensors are placed on the ear, which is often hairy. The sensors would lose their grip, causing the robot to spin in circles because it thought the user was wiggling their ear when they were just losing contact.

The Bottom Line

This paper proves that wiggling your ears can be a viable, socially invisible way to drive a wheelchair.

It's like discovering a hidden remote control button on your body that no one else can see. While the technology needs some polishing (especially dealing with ear hair!), it offers a future where people with severe disabilities can navigate the world without stopping their conversations or drawing attention to their disability. It turns a "useless" evolutionary leftover into a powerful tool for independence.

Technical Summary

1. Problem Statement

Current Human-Machine Interfaces (HMIs) for Electric Powered Wheelchairs (EPWs), such as joysticks, sip-and-puff, head movements, blinking, or tongue control, present significant limitations for users with neuromuscular disabilities (e.g., tetraplegia, muscular dystrophy, ALS).

• Social Stigma: Many existing methods (e.g., blinking, tongue movement) are intrusive, hindering social interaction and reducing user independence.
• Inadequacy: Surveys indicate that up to 40% of users struggle with current steering technologies.
• Gap: There is a lack of research on using Auricular Muscles (AM) (ear muscles) for control. These muscles are vestigial, controlled separately from the facial nerve, and located around the ear, offering a discreet, non-intrusive, and socially acceptable control interface that allows users to talk and move their heads freely.

2. Methodology

The study aimed to implement, compare, and optimize two distinct control strategies using surface Electromyography (sEMG) signals from the Auricular Muscles.

Hardware & Signal Acquisition:

• Sensor: MyoWare 2.0 Muscle Sensor (Instrumentation amplifier gain 200, bandpass 20–498 Hz).
• Signal Processing: The signal was rectified and passed through a Low Pass Filter (LPF) envelope detection circuit (cutoff 3.6 Hz).
• Sampling: Sampled at 47 Hz.
• Setup: A 1.8m x 3.0m wooden maze was used. A robotic prototype controlled via BLE protocol served as the wheelchair surrogate.
• Configuration: Continuous Control Strategy (CCS) used two wireless sensors (left/right ears); Morse Code Wheelchair Navigation (AM-MCWN) used one wired sensor.

Control Strategies:

1. Continuous Control Strategy (CCS):
   • Based on Schmalfuß et al. [16].
   • Logic: Constant contraction of the left ear turns right; right ear turns left; simultaneous contraction moves forward; no contraction stops. A double-quick contraction toggles "reverse mode."
   • Implementation: Uses a Finite State Machine (FSM) with a thresholding method to detect signal presence and duration.
2. Auricular Morse Code Wheelchair Navigation (AM-MCWN):
   • Based on Ka et al. [23].
   • Logic: Uses a single muscle group. Long/short signal combinations dictate movement (e.g., Long-Short = Left Turn; Short-Long = Reverse).
   • Implementation: A two-level SVM classifier. Level 1 detects signal presence; Level 2 classifies signal duration (Long vs. Short). Includes a 300ms/500ms buffer system to handle timing and a timeout mechanism for erroneous inputs.

Machine Learning & Classification:

• Classifier: Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel.
• Features: Time-domain features (Variance, RMS, Range, Kurtosis, Skewness, Slope changes) and Wavelet Transform Thresholds. Frequency domain features were excluded to reduce computational cost.
• Window Size Optimization: Various window lengths were tested. 300ms was identified as the optimal trade-off between accuracy and real-time responsiveness.
• Dataset: 30-second samples of mixed long, short, and noise signals were split into Training (80%), Validation (6.7%), and Testing (13.3%) sets.

Participants:

• Three healthy participants (aged 18–24), all capable of "ear-wiggling."
• Only one participant (P3) could move ears independently.
• Each participant completed two sessions (one for each strategy) after a 5–10 minute practice period.

3. Key Contributions

• First Direct Comparison: This is the first study to directly compare CCS and AM-MCWN strategies specifically using Auricular Muscles for wheelchair navigation.
• Optimization of Time Windows: The study empirically determined that while 300ms is generally optimal, the ideal window size varies by strategy (200ms for CCS, 300–500ms for AM-MCWN), a nuance previously unaddressed in literature.
• System Latency Reduction: Through the implementation of a two-level SVM and optimized buffering, the system delay was reduced from ~2000ms to approximately 200ms.
• Feasibility Validation: Demonstrated that vestigial muscles can be trained to control assistive devices with high accuracy (96% in SVM training) without hindering social interaction.

4. Results

Performance Metrics (Table 1):

• Time: Completion times varied significantly by participant and strategy. P3 (independent ear control) was fastest with AM-MCWN (4:00), while P1 was fastest with CCS (4:00).
• Crashes: CCS generally resulted in fewer crashes for P1 (3) but more for P2 (8). AM-MCWN showed consistent crash rates (5) across participants.
• Instructions: AM-MCWN required more distinct instructions (e.g., 122 for P1) compared to CCS (72 for P1), indicating a higher cognitive load for command generation.

User Feedback (Table 2 & Fatigue):

• Fatigue: Participants reported higher fatigue increases after using CCS (up to +2 points) compared to AM-MCWN (0–1 points), suggesting CCS requires more sustained muscle contraction.
• Usability:
  • CCS: Rated as more intuitive but harder to maintain steady signals. P3 (independent control) reported no fatigue increase.
  • AM-MCWN: Rated as more accurate and feasible for disabled users but more difficult to use due to the complexity of timing long/short signals.
• Navigation: Wireless connectivity issues caused occasional signal loss, resulting in the robot spinning. Users frequently triggered reverse mode unintentionally.

5. Significance and Conclusion

• Social & Functional Impact: The study confirms that AM control offers a "cosmetic" advantage, allowing users to control wheelchairs without disrupting speech or social cues, potentially improving the quality of life for tetraplegic users.
• Strategy Suitability:
  • AM-MCWN is better suited for users with limited motor control (e.g., stroke survivors) who can only control one muscle group, as it requires less sustained effort.
  • CCS is better suited for users with stronger motor skills (e.g., tetraplegics with some residual control) who can manage independent bilateral muscle activation, offering a more intuitive, continuous control feel.
• Future Work: The authors emphasize the need for:
  • A structured training program to help users master independent ear movement and signal timing.
  • Development of specialized electrodes to overcome adhesion issues caused by hair in the auricular region.
  • Testing on real powered wheelchairs rather than prototypes to validate safety and cognitive load in real-world environments.
  • Exploration of hybrid systems (combining AM-EMG with MMG, eye-tracking, or blinking) to increase accuracy and command diversity.

In conclusion, the paper establishes Auricular Muscles as a viable, socially acceptable, and technically feasible interface for powered wheelchair control, provided that user training and hardware connectivity challenges are addressed.