Laboratory of Harmonious Systems Engineering
Development of Balloon Robots Using Machine Learning
We are developing flying balloon robots in order to effectively utilize the space above attendees at concerts and other indoor events. We determine the position of the robot using multiple fixed cameras to image an infrared marker on the robot, and the resulting position is sent wirelessly to a microcontroller on the robot. The microcontroller combines this information with data from an onboard speed sensor, gyroscope and magnetometer to determine its position, velocity and direction. The microcontroller uses this information, together with a PID controller and propeller to minimize deviation from the desired course. We use neural networks to learn details regarding the specific balloon and effects of buoyancy and propeller output, allowing us to optimize the PID controller’s output. Our goal is to create a system that follows time and route constraints, within 10cm of error in large enclosed spaces with air currents.