Design and control of an autonomous electrical vehicle for indoor transport applications

As the need of intelligent vehicles on our roadways emerges, there is an equally important need emerges as well: The need of intelligent vehicles on areas such as university campuses, airports or shopping malls. These intelligent vehicles can help elderly, disabled, or people with heavy luggage. This thesis describes an intelligent vehicle that can be used indoor areas where pedestrians exist. The vehicle is planned to carry luggages and transport humans. Vehicle used is an electric golf cart, considering the significant advantages of less noise, no toxic gas emission and higher maneuverability. Firstly, vehicle is modified for unmanned drive. Drivers are added to control actuators on steering wheel and brake pedal. Then, main controller, dSpace MABX2 is placed. This device runs a MATLAB simulink model embedded in itself. While running, this model communicates with real world through input and output pins on the device, which are related to RTI blocks placed inside simulink model. Controllers are constructed in this simulink model and actuators were ready to control by connecting this in/out pins to related elements with cable. Other than the main controller, a separate controller, an Arduino board is used for braking, for emergency purposes. If an emergency situation occurs, if brake signal is cut off from main controller or if button on the related RC transmitter is pressed, this controller applies full braking independent of the main controller. PID controllers are preferred for steering wheel, brake and throttle unmanned drive subsystems. Indoor positioning is one of the most important problems when it comes to autonomous vehicles.There are studies proposing several computer vision based, wireless signal based etc. methods. Most accurate method is (IPS) but it is costly to set up and because of wireless signals gets weaker while passing through walls, it is not the best solution for every indoor environment. In this study, an encoder is used as main sensor for calculating position. Error caused by tire slip is very small because of the flat surface and slow move speed of the vehicle. But because of the error being accumulative, on long distance travel, real position and calculated position differ slightly. A computer vision based method similar to landmarking could be implemented in future phases to correct this difference. Environment identification and decision making is necessary for autonomous drive. For detecting obstacles and pedestrians in front of the vehicle, a LIDAR sensor is used. 3d cloud data consisting of 4 plane, can be obtained from this sensor. With the 4 plane LIDAR sensor used, it is possible to separate pedestrians from static obstacles and measure their movement speed. A second 1 plane scanning LIDAR with wide scan angle added to detect objects falling out of the 4 plane LIDAR scan angle, for the purpose of achieving more stable and safer obstacle avoidance. Avoiding obstacles is first priority for the vehicle. Some path following algorithms had been experimented on. General path following logic is based on goal points. To travel between two destinations in a known map, vehicle is given a number of goal points in proper order. Vehicle follows this points using implemented path following algorithm until the last goal point is reached. Last goal point means vehicle arrived the destination. On first experimental path following algorithm, vehicle calculates error of heading between itself and the goal point and rotates towards goal point by selecting the shortest direction, using a control logic. Moreover, vehicle constantly checks if the goal point is in vehicle's minimum turning radius. If it is, vehicle will never be able to reach it while trying to rotate towards it. Instead, vehicle maneuvers to opposite direction until the point is out of the minimum turning radius. Then rotates towards it. Second and final experiment is potential field method. A method including both path following and obstacle avoidance behaviors. Calculating pushing forces proportional to distances from objects in front of the vehicle and pulling force proportional to distance from next goal point, vehicle is able to maneuver between obstacles and reach the point. In this thesis, various stages of design and production of an autonomous vehicle, which is planned to operate in indoor environment where pedestrians exists, is explained. Sensors and mechatronic systems used for unmanned drive were presented, hardware and software used for control are discussed. Moreover, algorithms used for the vehicle to travel autonomously and and sensors used for receiving environmental data are explained. Finally, the real world driving tests performed are shown and the results were discussed.