The Thrust Vectoring Quadrotor (TVQuad) project arose from the need for fast, agile, and highly intelligent search and rescue robots in disaster scenarios. Aerial vehicles can achieve this goal safely and efficiently, but new technologies were needed to allow for heavier payloads and longer flight times. By employing thrust vectoring and variable pitch propeller technology, the TVQuad surpasses the size constraints limiting commercially available quadrotors, while maintaining much of their maneuverability and mechanical simplicity. By incorporating substantial onboard computing power, the TVQuad can also automate flight control and decision making processes, allowing rescue operators to focus on high level tasks like victim identification.
– Large payloads
– High maneuverability
– Autonomous motion planning controller
The Thrust Vectoring Quadrotor (TVQuad) project builds on the recent explosion of low-cost unmanned aerial vehicle (UAV) technology to create and control large vehicle sizes that were previously unattainable. Increased size allows the TVQuad greater capability in existing mission roles, new mission roles previously impossible, and longer flight endurance. Leading the project since it’s 2011 inception, I designed the Mk 1 and Mk 2 TVQuad prototypes, created a control system to overcome the TVQuad’s complexity, and developed a performance analysis technique capable of characterizing the control system’s abilities. Beyond design, I also handled 3D printing and other manufacturing tasks to construct the prototypes, liaised with suppliers, and remotely piloted UAV test flights.
I developed the TVQuad prototypes to demonstrate the new thrust vectoring technology at the core of the project. For the Mk 1 prototype, I designed a new 3D printed structure that allowed the propeller pods to pivot independently, but retained the direct drive fixed pitch design typical of commercial quadrotors. As the first quadrotor employing large angle thrust vectoring, the Mk 1 exhibited significantly improved maneuverability during concept flight tests in kSim 2.0.
The Mk 1’s thrust vectoring capability performed well in initial physical testing, but the large propeller size needed to lift the Mk 1 placed a great deal of strain on the main motor systems. This large strain slowed the Mk 1’s control response and stressed the speed controllers, inciting the rapid unscheduled disassembly of one controller during testing. While larger motors and controllers could resolve this issue at the expense of payload and flight time, I felt that a more elegant solution could be achieved by creating a Mk 2 prototype.
At 30 kg (65 lb) and 3 meters (9.5 ft) across, the Mk 2 prototype I am developing now is possible due to thrust vectoring rotor pods and variable pitch propellers. The thrust vectoring technology pioneered on the Mk 1 allows the Mk 2’s overall mass to be increased without significantly slowing its maneuverability. Similarly, switching from fixed pitch to variable pitch propellers allows the Mk 2 to use large propellers without slowing its control response. When completed, the Mk 2 promises improved flight time and efficiency with unprecedented lift capability.
Flight Controller Development
I designed the TVQuad’s nonlinear model predictive controller (NMPC) by leveraging my kinodynamic planning experience from previous work on the Cricket. Natively adept at handling complex vehicles and goals, the NMPC employs a novel combination of e-greedy tree-type planning and belief space modelling, coupled with efficiency improvements from an adaptive pre-controller. In preliminary testing, the controller has proven quite adept at maneuvering the TVQuad through confined simulated environments in Quadrotor mode.
The challenge in applying NMPC to aerial vehicles arises due to the limited time available for control decisions. On the TVQuad, the NMPC is required to produce a new set of control signals ten times per second, which restricts the number of available options the controller can analyze. As such, the performance and safety of the TVQuad rests on the consistency with which the NMPC identifies a “good enough” set of control signals within that tenth of a second.
As a key safety measure, I developed a performance analysis method to determine whether the TVQuad maintains a safe level of dynamic stability during a mission. The method performs a continuous Fourier analysis followed by a progressive linear regression to determine whether the vehicle’s flight profile is trending towards a user specified set of safe target behaviours. Testing the vehicle and controller under a wide range of conditions then allows one to determine a safe flight envelope for the vehicle.
With its implications on usability, my remaining work on the TVQuad project will be centred on understanding performance relationship between planning time and safe operation of the NMPC. On the hardware side, I am measuring the physical parameters needed for accurate simulation and control of the TVQuad Mk 2. In software, I am running flight tests of NMPC coupled to the TVQuad with thrust vectoring enabled. Depending on the controller’s performance in simulation, I have yet to decide whether physical flight tests will be safe.
– Performance Analysis of Nonlinear Model Predictive Control Applied to Multi-Rotor Unmanned Aerial Vehicles (pdf)
– Performance Analysis of an Enhanced Model Predictive Control for Safe, Semi-Autonomous Aerial Vehicle Flight in Confined Spaces (pdf)
– Performance Analysis of a Nonlinear Model Predictive Control for Safe, Robust and Semi-Autonomous Aerial Vehicle Flight in Confined Spaces (pdf)
– Rapid Control Selection through Hill-Climbing Methods (official) (pdf)
– 2015 International Conference on Robotics and Automation (ICRA2015) PhD Forum
– 2015 University of Calgary Mechanical Engineering Graduate Conference (video)
– 2014 University of Calgary Mechanical Engineering Graduate Conference
– 2013 University of Calgary Mechanical Engineering Graduate Conference