SCIENCE AND ENGINEERING FAIR
Research Plan and/or Abstract for 2019

Category Award:   0 (GRAND AWARD)

Research Plan:
I conducted research into the use of LiDAR for developing collision-avoidance algorithms. With the rise in autonomous aircraft, collision-avoidance is a major hurdle to its deeper integration into our lives. My research sought to explore the use of LiDAR sensor technology for the perception of obstacles, as well as efficient real-time pathfinding methods using this technology. Other depth sensors only permit the perception of average distance to an obstacle within the sensor’s field of view, while LiDAR sensors allow for the rapid perception of both the depth and location of obstacles. The LiDAR returns points in 3D space in relation to the sensors representing obstacles that interrupted the path of the laser. The resultant “point cloud” is a 3D scan of the environment. My research involved the use of this technology within a computer-simulated flight environment. Within this environment, I was able to create a new collision-avoidance algorithm leveraging this simulated LiDAR data. The algorithm works by first processing the point cloud to create a “depth image negative,” or map of the clear area around the aircraft, whereby the aircraft is represented by a rectangular prism. Based on the velocity of the aircraft, the location of the prism is placed within the depth image negative, showing the predicted flight path. If the prism infringes on an area not within clear air, the aircraft is signaled to alter its course.

Abstract:
Navigating a UAV in complex and dynamic environments poses challenges for human operators, and can be mitigated through the use of autonomous control systems. In this paper, a novel method for the autonomous navigation of a UAV using a 3D perception sensor is considered. In our application, LiDAR (Light imaging, Detection and Ranging), a sensor that perceives objects in the environment through laser light pulses, measures both obstacle location in addition to distance. A new algorithm that takes into account the known kinematics of the UAV and the LiDAR data is developed to create a high-performance collision avoidance algorithm. The algorithm relies on predicting the trajectory of the UAV and extrapolating data from a LiDAR point cloud to determine if a collision is imminent, in which case the algorithm calculates a new trajectory to a clear area. In experiments, the algorithm was 20-40 times faster than current pathfinding methods. It accomplished this by calculating only a single path between two nodes each iteration rather than an entire route, but at the expense of routing efficiency. This behavior lead the UAV to be more capable of autonomous navigation than current methods in dynamically changing environments.