Design and computational evaluation of an eye gaze wheelchair retrofit with temporal command supervision

For millions of people living with severe physical disabilities, the simple act of moving from one room to another can be a monumental challenge. While powered wheelchairs offer a path to independence, they rely on a control interface that the user must be able to operate, typically a joystick requiring hand strength and dexterity. When hands cannot function, the eyes often remain a powerful, untapped resource. The human eye is naturally precise, capable of locking onto objects with remarkable speed, and for those who can see but cannot move their limbs, looking at a direction can become a command to move. However, turning a glance into safe, reliable motion is not as simple as pointing and clicking. The eyes are constantly scanning the world, shifting focus to read a sign or watch a person, and a control system that moves the chair every time the eyes land on a spot would be chaotic and dangerous. The core challenge for engineers is to distinguish between a deliberate intention to move and a fleeting glance, and to ensure that if the connection between the user's mind and the machine breaks, the chair stops immediately rather than continuing on a dangerous path.

Researchers at the Kuwait College of Science and Technology have tackled this problem by building and testing a system called iMove, a retrofit kit that allows a standard powered wheelchair to be controlled entirely by eye gaze. The team, led by Hasan Fayyad-Kazan, constructed a device that sits between a commercial eye-tracking camera and the wheelchair's existing joystick. Instead of the user moving a stick, they look at a specific direction on a tablet screen, and a small robotic arm attached to the wheelchair physically pushes the joystick in that direction. The system includes a central "STOP" target and eight directional targets. When a user looks at a direction, the system waits to see if the gaze is steady enough to be an intentional command. If the gaze holds long enough, the robotic arm moves the joystick, and the wheelchair begins to roll. If the user looks away, the system is designed to stop the motion.

The researchers began by testing a basic version of this system to see if it could translate a look into a movement. In a preliminary trial, they attempted to select a direction ten times. The system successfully moved the joystick eight times, achieving an 80 percent success rate. However, this initial test revealed a critical flaw in the basic design: the system was too trusting. If the user's eyes drifted off a target, or if the computer sending the signal froze, the basic system would keep the joystick pushed and the wheelchair moving. It did not know how to stop on its own when the input became invalid. This is a dangerous scenario, as a wheelchair that continues to move after the user has lost control or lost focus could lead to a collision.

To fix this, the team developed a "temporal supervisor," a layer of software logic that acts as a safety guard for the commands. This supervisor introduces a set of rules to verify that a command is still valid before allowing the chair to move. First, it requires a "dwell" time, meaning the user must look at a target for a specific duration—set at 0.60 seconds in their design—before the chair moves. This prevents accidental movements caused by a quick glance. Second, the system checks if the input is "fresh," ensuring that the signal coming from the eye tracker is new and not a frozen image from a moment ago. Third, it monitors the connection between the tablet and the wheelchair; if the communication link breaks, the system immediately cuts the command.

The researchers did not just build this system; they rigorously simulated thousands of scenarios to see how it would behave under stress. They created a digital environment where they could inject faults, such as freezing the eye tracker, cutting the communication cable, or making the computer stop processing. In these simulations, the basic system failed every time a fault occurred, keeping the wheelchair moving for the entire duration of the test. In contrast, the new supervisor system reacted quickly. When the eye tracker froze, the supervisor stopped the command within 0.12 seconds. When the communication link was cut, it stopped the command within 0.24 seconds. The team also tested what would happen if a user accidentally looked at a target for a very brief moment, such as 0.12 seconds. The basic system would have activated the chair, but the supervisor correctly ignored it because the gaze was too short to be a deliberate order.

The study also looked at the physical reality of stopping. Even if the software sends a "stop" command instantly, the wheelchair does not halt in mid-air. It has momentum. Using the observed stopping time of the wheelchair, which took between 3 and 4 seconds to come to a complete rest, the researchers calculated how far the chair would travel after a stop command was issued. At a speed of 0.50 meters per second, the chair would travel between 0.78 and 1.13 meters before stopping completely. This distance is significant, highlighting that the speed of the software is only one part of the safety equation; the physical braking of the chair is equally important. The team found that if the system waited too long to issue a stop command, the chair would travel even further, emphasizing the need for the rapid response times provided by their supervisor.

While the results are promising, the researchers are careful to note that this work is a proof of concept rather than a finished medical product. The testing was done by the developers themselves, not by the people who would eventually use the chair, and the success rate of 80 percent was based on a very small number of attempts. The study proves that the logic works in a computer simulation and that the hardware can physically move the joystick, but it does not yet prove that a real user can navigate a complex environment safely. The team emphasizes that future work must involve testing with actual users to see how they interact with the system over longer periods and in real-world conditions. They also note that the system needs to be robust against mechanical failures, such as the robotic arm getting stuck, which the current software cannot detect.

Ultimately, this research highlights a vital principle in assistive technology: safety is not just about making a device work, but about making sure it stops when it should. By combining a deliberate selection method with a smart safety layer that watches for errors, the researchers have created a blueprint for a safer way to control a wheelchair with the eyes. The iMove system demonstrates that with the right combination of timing checks and fault detection, it is possible to turn the natural, fluid movement of the human gaze into a reliable and safe command for a machine, bridging the gap between intention and action for those who need it most.