Control of a semi-autonomous powerchair using head movements

Wheelchairs are crucial for the transportation needs of the people whose ability to move is limited. Wheelchair-bound people, who have enough muscle strength on their arms for controlling the wheels of the wheelchair, could easily use the manual wheelchairs. However, there are also wheelchair-bound people who may not have enough strength to control the wheelchair manually. This type of people generally prefers to use powerchairs. The most basic components of powerchairs are a joystick for the user to control the chair and some number of motors driving the wheels. By manipulating the joystick, the user provides references of angular and linear velocities for the powerchair to accelerate. After user manipulation, the velocity reference values are transferred to the motors that are responsible for driving the wheels. There are also some people who do not even have the ability to use the joystick to control the powerchair. For such people, there is a need for other types of control mechanism, which involves an assistive system. In this study, a powerchair is developed for the paraplegic or partially paralyzed people. The developed powerchair is designed to be controlled by the head movements of the user instead of a joystick. The orientation sensor (IMU) is used to detect the head movements of the user and give angular and linear velocity reference values to the system. While orientation changes on the X axis give the linear velocity refernce value, orientation changes on the Y axis give the angular velocity reference value. While the powerchair can be controlled more sensitively using a joystick, it is more difficult to control the chair with head movements. To overcome this issue and provide the safety for the powerchair and the user, additional systems are needed. The developed system presents a semi-autonomous obstacle avoidance method to increase the level of safety during the drive. In the scope of this thesis, a FGM based method called FGM-I is developed as a semi-autonomous obstacle avoidance algorithm. FGM provides a safe transfer to the goal point for the vehicle by using the largest gaps in the field of view and calculates final heading angle. Although the selection of the largest gap yields better results in terms of safety, sometimes the trajectory could be longer than necessary. In addition, when the differences between gap size values are too small, the robot switch from one gap to another and this sometimes causes traveling too close or even hit the obstacles during the movement due to change in gap selection caused by the small differences between the gap sizes. On the other hand, FGM-I method chooses the gap according to the length of the gap and the angle between the middle point of the gap and the target point using a new utility function. In addition, the coefficients determine how much each parameter affects the utility function. These coefficients can be chosen to give more importance on either the gap size or the angle between the gaps and the goal point. An additional reward constant is added to the selected gap's utility value and this yields a penalty effect on changing the selected gap. On the other hand, in order to make a fair comparison, 100 Monte Carlo simulations in which the coordinates of the obstacles are specified randomly, are implemented for FGM-I and FGM. A LIDAR (Light Detection and Ranging) sensor is used in simulations to detect the obstacles with a total 180 degree field of view and 6 m range. In these simulations, 10 rectangular obstacles with various size and positions are distributed randomly to the 35X25m environment. Initial and goal coordinates are given as [0-0] and [35-25] respectively . The results are compared in terms of safety, length of path and zigzag density. According to the results, the FGM-I generates 14% shorter trajectories than the FGM, while the average safety metric is only 4% worse. In addition, the average yaw-rate metric of FGM-I is 16% lower than the FGM. This means that the resulted trajectories are smoother and zigzag problem of FGM is decreased in a significant rate. The user of the powerchair provides the target angle by head movements to be used as an input of the FGM-I, as well as for the calculation of final orientation angle. At the end, a semi-autonomous controllable system is developed for the people to move to a goal point with the powerchair without any collision. The ability of providing linear velocity to the system by head movements may also result in dangerous situations, which reveals the need for controlling the velocity given by the user. In order to provide a solution for this requirement, a fuzzy logic based velocity planning method is integrated to the system. In this method, a risk factor is calculated by fuzzy logic operation based on the distances to obstacles and position of the obstacles. This generates a risk factor according to environmental factors. Using the environmental risk factor and the linear velocity reference provided by the user's head movement, dynamic risk factor is calculated by fuzzy logic. A safe velocity reference was obtained with this dynamic risk factor. Velocity reference calculated by the fuzzy logic operations and the velocity reference which is given to the system by the user are weighted by the coefficients calculated by using the dynamic risk factors. If the dynamic risk factor is large, the weight of the safe velocity reference is high. If the dynamic risk factor is small, the weight of the velocity reference which is given to the system by the user is high. By doing so, a safe linear velocity reference to be used in the movement of the system is obtained semi-autonomously. Semi-autonomous velocity planner and FGM-I were tested together on Gazebo simulation environment by users. 20 users tested the semi-autonomous powerchair using their head movements. In these simulations, two different velocity planner and without velocity planner systems were tested, while the FGM-I is used for obstacle avoidance. According to the simulation results, average safety metric is 16% better than without planner system. This means that safer trajectories are generated for users and powerchair by using the dynamic velocity planner. In addition to simulations, obstacle avoidance and velocity planner systems were tested together on experimental setup. At this point, the powerchair has been semi-autonomously controlled by head movements with both obstacle avoidance and velocity planning support.