Multiphysics-based Autonomous Energy-Optimal Planning and Control of Multirotor Unmanned Aerial Vehicle (ONR White Paper was accepted but no tracking number was provided.)
Abstract
The goal of this project is to formulate a novel multiphysics-based framework for integrated unmanned aerial vehicle (UAV) motion planning and control, and enable autonomous UAV operation for enhanced energy performance with the aid of machine learning. In our previous 2-year NEPTUNE 1.0 project, we have successfully developed and validated a multiphysical UAV model, which captures the rotor aerodynamics, motor and motor controller electro-mechanical dynamics, battery electrical dynamics, and frame rigid-body dynamics. To the best of our knowledge, this model is the first-of-its-kind in capturing all these relevant dynamics, and more importantly, their coupling during UAV operation. Based on the model, we have performed optimal UAV motion planning for basic maneuvers, e.g. straight-line hover-to- hover operation. It is inspiring to find (currently in simulation) that even for such simple maneuver, the multiphysics-informed approach could achieve 6-25% improvement in energy assumption compared with baseline. Besides these exciting findings, we have identified new critical topics to expand and materialize our new approach. The first task of this project is to establish a framework that integrates high-level motion planning with low-level control. The proposed motion planning scheme uses the actual onboard actuation, i.e. PWM command to each rotor, as the decision variables for optimization. While such practice unleashes the full potential of energy saving by exploiting the complete multiphysical dynamics, the obtained open-loop PWM command sequence is not suitable for direct real-time control due to the commonly presented disturbances and stability issues. We plan to integrate the motion planning with real-time feedback control, which regulates certain states of the UAV based on sensor feedback to realize the planned maneuver. Such high-level planning plus low-level control architecture has been a common practice in UAV control, but our new multiphysics-based approach requires a reformulation of the architecture including redefining the planning and control variables, and redesigning the low-level controller. The second task of this project is to apply the new framework for autonomous energy-efficient generic and special UAV operation in real time. First, we will develop the capability for autonomous real-time motion planning to replace the current computation-heavy offline optimization routine. This part of the research would involve establishing a library of planning sequences for different maneuvers, extracting features for different maneuver classes, and using machine-learning methods to train a real-time planning module capable of generating the motion sequence autonomously. Second, we are also interested in enabling novel UAV operation modes. One example is partial-rotor operation, where some rotors are shut off under certain operating and load conditions. Such operating mode has demonstrated energy saving potential, but would require more sophisticated control to ensure safety and reliability, which can be enabled by the proposed multiphysics-based planning and control framework. The proposed research will be collaborated with and supported by the Naval Information Warfare Center (NIWC) Pacific. We will also continue our successful experience of mentoring military students. In our NEPTUNE 1.0 project, we have trained 1 ROTC student (among others) on UAV control and testing, who is now an avionic testing engineer with the US Air Force.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 15, 2021
- Source ID
- N000142112080
Entities
People
- Xinfan Lin
Organizations
- Office of Naval Research
- United States Navy
- University of California, Davis