Event
MSSE Thesis Defense: Swapneel Naphade, "Estimation and Control of Autonomous Racing Drone"
Wednesday, April 15, 2020
2:00 p.m.
WebEx online meeting
Swapneel Naphade
naphadeswapneel@gmail.com
https://naphadeswapneel.my.webex.com/naphadeswapneel.my/j.php?MTID=m19f8f29082768e99d8c9ce24b40634e6
MSSE Thesis Defense
Estimation and Control of Autonomous Racing Drone
Swapneel Naphade
Autonomous Drone Racing (ADR) is an annual competition, organized at the International Conference on Intelligent Robots and Systems (IROS), in which research groups all over the world participate to demonstrate the state-of-the-art technology in the autonomous aerial robotics field. The challenge in the competition is to develop an autonomous aerial robotic system, capable of traversing a known drone racing course, without collisions, using only the onboard sensing and computing resources. Autonomous Drone Racing demands high precision and speed in estimation and control of the racing drones for successful completion of racing laps.
This work describes the system development of the Autonomous Racing Drone System for the IROS ADR competition. A gate detection based, computationally light-weight visual-inertial localization (VIL) system which fuses information from vision, optical flow and IMU sensors for race-course gate position estimation is developed. The VIL system utilizes a Linear Time-Variant Kalman Filter which consists of asynchronously executed prediction and measurement steps. We show that this method provides better gate position estimation even in the presence of noisy and intermittent visual gate measurements. We also show that the proposed VIL system has a significantly lower memory usage than the state-of-the-art Monocular VIO systems which makes the proposed system suitable to run on resource constraint hardware. A non-linear model predictive control (NMPC) strategy is implemented for high-speed way-point navigation of the racing drone. NMPC enables controlled aggressive maneuvers of the drone with a facility to limit the control inputs which makes the system more intuitive to tune in a short period. We show that the NMPC strategy provides better trajectory tracking performance as compared with the traditional PD controller.
The VIL system proposed in this work was utilized in the autonomous drone racing system which won the second-place in the IROS ADR 2019, Macau competition.