Game of drones

• Main Author: Wei, LuoBin
• Other Authors: Huang Shell Ying
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/148044
• Institution: Nanyang Technological University

This report proposes a solution to the Tier 1 task of a simulation-based drone racing event organised by Microsoft Research (Game of Drones). Objective of Tier 1 task of Game of Drones was to fly a drone through a series of gates with the shortest possible lap time while avoiding any collision with a competing drone or environmental obstacles without utilizing computer vision for navigation. The proposed solution is executed in a modified AirSim simulator, powered by Unreal Engine4. Application-Programming-Interface(API) is provided and utilised by race organisers. Race Participants are expected to implement their proposed solution via the APIs provided in the simulator. The solution proposed by this report, to achieve the shortest possible lap time, focus on developing the following area: Trajectory Planning, Trajectory Tracking and Collision Avoidance Algorithm. Efficient Trajectory Planning is achieved by generating a Cubic Hermite Spline trajectory utilizing gate positions as data-points for interpolation and a vector pointing to the subsequent gate as the constraint. Additional data-points are dynamically added if needed for a more accurate trajectory when utilizing vectors as constraint. Augmentation of data-points within gate position via gate segmentation for a more effective trajectory planning is also explored. Proportional-Integral-Derivative (PID) and Pure Pursuit controller are utilised to obtain an accurate trajectory tracking. Reactive collision avoidance is implemented by replanning trajectories when detecting an opponent. Game of Drones2019 concluded by the commencement of this project. Leaderboard solution reports published by Game of Drones organisers depicts a majority of the solutions relied/required manual static tweaks to the specific racing environment to achieve a competitive short lap time. This report will aim to solve Tier 1 dynamically, with minimal tweaks to the racing environment while maintaining a competitive lap time. Therefore acquiring a solution that would work in any racing configuration and not specific to a singular environment. Results of this report would be compared against the Game of Drones2019 Leaderboard result .