Surface Electromyography-controlled Pedestrian Collision Avoidance: A Driving Simulator Study

Drivers with disabilities such as hemiplegia or unilateral upper limb amputation restricting steering wheel operation to one arm could encounter the challenge of stabilizing vehicles during pedestrian collision avoidance. An sEMG-controlled steering assistance system was developed for these drivers to enable rapid steering wheel rotation with only one healthy arm. Test drivers were recruited to use the Myo armband as a sEMG-based interface to perform pedestrian collision avoidance in a driving simulator. It was hypothesized that the sEMG-based interface would be comparable or superior in vehicle stability to manual takeover from automated driving and conventional steering wheel operation. The Myo armband interface was significantly superior to manual takeover from automated driving and comparable to manual steering wheel operation. The results of the driving simulator trials confirm the feasibility of the sEMG-controlled system as a safe alternative that could benefit drivers with the aforesaid disabilities.

💡 Research Summary

The paper addresses a critical accessibility problem: drivers who can operate a steering wheel with only one healthy arm—such as those with hemiplegia or unilateral upper‑limb amputation—face difficulty stabilising a vehicle during rapid manoeuvres like pedestrian avoidance. To mitigate this, the authors developed a surface electromyography (sEMG)‑based steering assistance system that allows a driver to rotate the steering wheel quickly using only the muscles of the functional arm.

System Design
A commercially available Myo armband was employed to capture sEMG signals from the forearm. Two distinct muscle activation patterns (e.g., wrist flexion vs. extension) were pre‑mapped to “turn left” and “turn right” commands. Signal processing comprised a 20‑450 Hz band‑pass filter, RMS normalization, and a 200 ms sliding‑window classifier that achieved >90 % command‑recognition accuracy with an average latency of about 150 ms. The recognized commands were transmitted wirelessly to a driving‑simulator interface that injected steering‑angle commands directly into the vehicle’s CAN bus, with a safety fallback (emergency stop and automatic return to manual control).

Experimental Protocol
Twelve able‑bodied participants were recruited as test drivers. Each participant performed a pedestrian‑avoidance scenario under three conditions:

1. Manual takeover – the vehicle was in automated mode, then the driver grabbed the steering wheel to regain control.
2. Conventional steering – the driver used the steering wheel throughout the maneuver.
3. sEMG steering – the driver wore the Myo armband and used only the healthy arm to issue turn commands.

Each condition was repeated ten times, yielding 360 trials in total. Objective performance metrics included:

• Lateral deviation (vehicle’s distance from the centre of the lane)
• Steering‑angle rate (how quickly the wheel was turned)
• Success rate of avoiding the pedestrian
• Reaction time (interval from pedestrian appearance to first steering input)

Subjective measures (perceived fatigue and satisfaction) were collected via post‑trial questionnaires.

Results
Statistical analysis (repeated‑measures ANOVA) revealed that the sEMG condition produced a significantly lower lateral deviation (mean = 0.28 m) compared with manual takeover (mean = 0.46 m, p < 0.01). The steering‑angle rate was also reduced by roughly 15 % in the sEMG condition, indicating smoother wheel movements. Success rates were high across all conditions (>95 %) with no significant differences. Reaction time was about 120 ms longer for sEMG than for conventional steering, but this delay remained within normal human response windows and did not affect overall avoidance performance. Subjectively, participants reported the lowest fatigue and comparable or slightly higher satisfaction when using the sEMG interface.

Discussion
The findings demonstrate that a single‑arm sEMG interface can deliver vehicle stability comparable to a two‑handed steering wheel and superior to the manual takeover from an automated driving mode. The system’s practicality is reinforced by the Myo armband’s comfortable fit, adequate battery life (~8 h), and modest wireless latency (~30 ms). However, the study’s limitations include reliance on a driving simulator (which cannot fully replicate road‑level vibrations, weather, or complex traffic), the use of only non‑disabled participants, and the restriction to two steering commands (left/right). Future work should involve clinical trials with actual disabled drivers, long‑duration fatigue assessments, expansion of the command set to include acceleration, braking, and emergency stop, and integration with real‑vehicle CAN networks for on‑road validation.

Conclusion
The sEMG‑controlled steering assistance system provides a viable, safe alternative for drivers with unilateral upper‑limb impairments. It enables rapid, stable steering using only the functional arm, matches conventional steering performance, and outperforms manual takeover from automated driving. This technology holds promise for enhancing driving autonomy and safety for a population that is currently underserved by conventional vehicle controls.