Intelligent Unmanned Aerial Systems for Atmospheric Turbulence Estimation

Control Theory; Atmospheric Sciences; Multi-Vehicle Systems and Air Traffic Control

This project explores the capabilities of multirotors in replacing the balloon and tower systems for turbulence measurement, by using advanced atmospheric estimation, flight control, path planning.

Accurate measurements of ABL turbulence are critically important for ecological and atmospheric calculations, including net ecosystem exchange and fluxes of heat and momentum near Earth's surface. Additionally, the accuracy of weather forecasting models depends in part on precise representation of ABL turbulence. The proposed methods  will allow for ABL turbulence measurement using UAVs for the first time. The UAV measurement platform has the capacity to follow a turbulent eddy and track its development as it moves through the atmosphere. This would be a vast improvement over a stationary meteorological tower, which is only able to observe the passing of such eddies.

The proposed project brings together an interdisciplinary team from Electrical Engineering, Environmental Sciences and Systems Engineering with research experience in adaptive control theory, boundary layer meteorology and intelligent motion planning. ​​By crossing school boundaries, the team can leverage expertise from the ​ ​College of Arts and Sciences and ​ the School of Engineering and Applied Sciences to explore a cross-cutting problem.​ The union between theoretical design and experimental application is central to the proposed research.  The collaborative effort of the team will not only present each the members an opportunity to advance individual research goal but also gain a better perspective in the larger field.​

A. Research Motivations

The atmospheric boundary layer (ABL) is the region of the atmosphere nearest to the surface of Earth. The height of the ABL, measured from the surface of the earth outwards, can vary considerably from tens of meters to several kilometers throughout the day-night cycle. Observing and forecasting weather in the ABL is central to many aspects of daily human life, including agriculture, aviation, and public health. Knowledge of wind conditions in the ABL is especially desirable: macro-scale winds drive aerosol transport through the atmosphere, while turbulence, random fluctuations in horizontal and vertical winds, controls the flux of heat and distribution of energy between Earth's surface and the atmosphere. Turbulence is organized into structures called eddies, which can cover a broad range of spatial and temporal scales. The smallest eddies present in the ABL are on the order of centimeters, and last for only several seconds. For this reason, high-frequency measurements (10-20 measurements per second) are required to observe turbulent eddies in the ABL.

Traditional methods for observing ABL winds utilize sensors attached to stationary towers or to tethered helium-filled balloons that can be raised and lowered through the atmosphere. Modern remote-sensing technology, including radar wind profilers and Doppler lidar instruments, can also observe ABL winds. However, all of these approaches are either require time-consuming assembly or are insufficient to probe the entire depth of the daytime ABL. Additionally, sensors capable of sampling at the high frequency required for turbulence measurements often cost tens of thousands of dollars. 

Recently, UAVs have proven to be reliable, inexpensive alternatives to traditional wind-sensing instruments for observing macro-scale winds. The proposed project seeks to extend the existing results  by improving UAV flight controllers to allow for measurement of ABL turbulence in addition to macro-scale winds.

B. Issues with Current  Techniques

Though the multirotor market is booming now, commonly anticipated performance goals for these systems have not been fully achieved, largely due to the  intrinsic drawbacks of the current  methods for rotor control:

  • Inadequate Treatment for  System Faults: Multirotors often work in complicated and severe environments and face the threat of actuator failures, as such unpredicted circumstances may cause physical damage to a rotor. Moreover, the dust and dampness that accompanies harsh working conditions may lead to the burning of motors. If the multirotor is not equipped with a proper control strategy, it may fail to complete its task when there is emergency.  However, the study of fault-tolerant control for multirotor systems has not been systematically undertaken by the research community. The widely used  PID controllers can only work with constant reference signals and cannot deal with high order dynamic systems like multirotors at other fast maneuver conditions (that is, a PID controller is not able to ensure the rotor system  track an arbitrary trajectory signal). Moreover, closed-loop system stability is not guaranteed by a PID controller applied to multirotor systems (which is only analytically ensured to stabilize low-order systems). 
  • Insufficient Accommodation of Center of Gravity Variations: For the tracking designs for four outputs,  most of the existing literature  assumes known control allocation schemes. Some adaptive control allocations have been presented in the literature, which assume that there only exist multiplication parameter uncertainties but no addition parameter uncertainties.  Such assumptions may not be appropriate for multirotors equipped with manipulators. The extension and flexion of the mechanical arm may cause shifting of the system mass center, which leads to uncertain control allocation scheme. If the multirotor delivers a cargo or picks up something, the mass and moments of inertia may change.  For systems with time-varying or uncertain parameters, PID control cannot guarantee either system stability or desired tracking performance. 
  • Unable to Adjust Control Objective Autonomously: A multirotor UAV may partially fulfill its task even if fatal actuator failures happen.  Existing downgrade designs are based on the assumption that failures can be detected accurately in very short time so that the motion planning algorithm can modify the desired trajectory to make the UAV land safely. However, it is very difficult to detect loss-of-control failures (where the thrust force generated at a failed rotor is random and unknown) for a multirotor under parameter uncertainties. Current control techniques are not able to aid the decision making in the part of motion planning. Learning algorithms are needed to update the mission in time to prevent a crash.

C. Research Problems

The primary method to estimate wind using a multirotor UAV  has several key limitations: first, the UAV must be set to a hover flight mode, and is unable to move horizontally. This prevents the UAV from tracking a turbulent eddy as it moves through the atmosphere. Second, the PID flight controller does not provide the high-frequency data necessary to measure turbulence. Finally, several parameters related to the drag coefficient of the UAV must be estimated in order to extract the wind information. This parameter estimation reduces the overall accuracy of the wind measurements. To improve these aspects of wind estimation, we aim to solve the given research problems: 

  1. Control framework for wind facing maneuver: In order to improve the measurement accuracy, the mounted sensor should always face the turbulence flow. However, the direction of the turbulence is unknown and changing fast. Proper control scheme is needed to guarantee the stability of the system and the accuracy of measurement.
  2. Intelligent  motion planning and flight strategy generation based on collected data: Limited by the mobility, the tower and balloon systems are placed at preselected locations based on comprehensive considerations of terrain, vegetation and buildings. The multirotor UAVs can fly to spots with more meaningful data if intelligent motion planning algorithm is implemented to the system.
  3. Higher accuracy estimation for  wind  speed: Literature shows that the indirect method has better performance in wind direction estimation than wind speed but the direct method has advantage in speed measuring. Advance data processing method is needed to improve the total accuracy.

D. Proposed Methods

To improve upon the shortcomings of the PID flight controller, the proposed project will utilize existing advanced control methods for multirotor systems, such as multivariate adaptive control and nonlinear back-stepping control.  We will also work on innovative ways of collecting data, processing data, and improving data collection. Path planning algorithms will be developed for agents in a multi-agent and distributed system under an uncertain environment and changing conditions, such as an agent experiencing a fault or newly restricted air space. The following methods will be investigated and utilized in the proposed research:

  1. Active tracking controller for passively wind direction following: We  propose an active control method to achieve the goal of wind facing  by proper control design rather than generating a reference signal by guidance algorithm. Such approach is also considered passive for the wind direction is not known before measured; the behavior of ``following'' is driven by the dynamic rules.
  2. Reinforcement learning based hierarchical motion planning: We propose an approach based on the classic potential field method. The wind field will be used as potential field to generate flight trajectory. In order to collect more information during a flight and capture the turbulence, the collected data will be used to update and predict the wind field. The prediction algorithm will be improved by reinforcement learning for more efficient data collecting.
  3. High accuracy estimation based on reduced-order extended Kalman filtering: Kalman filter is a well-established  estimation method for data in a dynamic system. We will  undertake further processing and information fusion of data collected by both direct and indirect methods to improve the results.

E. Research Goals and Timeline for the Project

The research goals are defined and will be achieved by fulfilling the some key research tasks for developing solutions to the stated research problems.

  • Task 1: Solution to Problem 1, and completion of two papers (2 months). The goal of this research is to develop advanced control  techniques for unmanned aerial systems. The control scheme will guarantee the safety of the system under uncertainties, faults and failures.
  • Task 2: Setting up a multirotor control lab (one quadrotor and one hexarotor have been purchased with some existing available funding, and control system implementation and flight test are the next tasks) (2 months). The goal of this period is to make essential preparation for the experiments and measurements of later tasks.
  • Task 3: Solution to Problem 2 and completion of one paper (2 months). The goal of this research is to develop novel motion planning algorithms to collect data more effectively.
  • Task 4: Writing 2-3 research proposals (3 months). The goal of this period is to seek external funding opportunities.
  • Task 5: Solution to Problem 3 and completion of one paper (3 months). The goal of this research is to implement the proposed control and planning approaches to the multirotors for more accurate data. The data set should also provide insight in the study of atmospheric turbulence structure.

F. Summary

The proposed project will explore the capabilities of the multirotor systems in replacing the current balloon systems and tower to meet the challenge of turbulence measurement in the lower atmosphere. The main technical idea is that more sophisticated multirotor flight maneuvers help improve the data collection efficiency and accuracy and more effective control systems are needed to realize the desired maneuvers. The research tasks are to study the integration of direct and indirect estimation methods for more accurate results and the desired multirotor flight strategy and control system design for realizing the desired flight trajectories. Rigorous methods that fuse both adaptive control and learning-based motion planning will be investigated to enhance the efficiency and accuracy of data collecting and processing. The control designs for multirotors with uncertainties and faults and the learning-based task switching algorithms obtained in this project will be the basis for proposals for resilient control not only in multirotor systems but also in a broad range of other performance-critical cyber-physical systems and smart and autonomous systems.

Desired outcomes

If successful, the proposed work will enable performance benefits promised by smart, adaptive, self-aware atmospheric measuring systems. The expected deliverables include high quality peer-reviewed papers and proposals submitted to external agencies. 

A. Publication Plan

During this project period, we will solve some fundamental tasks from the listed problems and then document the research results for some technical papers:

  • Paper 1: Adaptive Control Schemes for Underactuated Multirotor Systems under Actuator Failures and Wind Disturbances, which deals with the disturbance rejection and  failure compensation problem for  multirotor systems that can  generate thrust forces along one direction of its body frame (to be submitted to the 2019 AIAA Guidance, Navigation, and Control Conference).
  • Paper 2: Maneuverability-Based Path  Reconfiguration for Hexarotors Under Severe Actuator Failures}, which studies the control strategies for  hexarotor systems encountered with actuator failure. A fault detection algorithm will be designed for the system for controller switching to track four outputs for desired  performance. The stability of the switched system will be investigated and guaranteed by an adaptive controller (we plan to submit it to the 2019 IEEE Conference on Decision and Control).
  • Paper 3: Wind and turbulence estimation from multi-rotor aircraft using direct and indirect methods. This paper will address the challenges related to the estimate of wind and turbulence using multi-rotor aircraft and provides initial comparisons of the estimates with tower-based measurements.
  • Paper 4: Reinforcement learning based hierarchical motion planning,  which studies prediction algorithms based on reinforcement learning to plan trajectories. This approach will converge to an optimal policy, given enough data and trials (we plan to submit it to the 2019 IEEE Conference on Robotics and Automation).

B. Targets for External Funding

This seed funding will help us to obtain the preliminary needed for formulating impressive framework of advanced  techniques for the atmospheric measurement using multirotor systems and to write attractive proposals for external funding. Our group has expertise in adaptive control, motion planning and boundary layer processes, but needs more background and more results related to multirotor system modeling and control, to make us more competitive in this specialized area of application. This effort will also encourage and support our research collaboration between the School of Engineering and Applied Sciences and the College of Arts and Sciences.

  • ULI Program from NASA

We are planning to work with Aurora Flight Sciences on a proposal for the University Leadership Initiative (ULI) program of NASA. Safety, or assurance, is an area of long-term concern by NASA. Specifically, we propose to the topic of Complex Autonomous Systems Assurance (CASA) in the  ARMD Technical Challenge Statements  of NASA. The collaborative research between the company and us will lead to development and demonstration of innovative V&V tools and methods to provide assurance of the safe operation of complex, increasingly autonomous, non-deterministic systems. 

  • Cyber-Physical Systems (CPS) Program from NSF

For the 2019 cycle, a NSF CPS proposal is being planned to investigate issues in control, autonomy, and resilience and safety for  multirotorcontrol systems technology. UAVs  often work in spaces with other robots and humans, so they have to learn to behave in a safe way when emergencies happen.   Learning algorithms are needed to update the mission in the event of platform faults. When the mission requires high-precision  tracking,  the guidance algorithm should be controller-aware to guarantee performance. The guidance system should also be dynamics-aware for parameter uncertainty and safety-aware for failures and damages. A key aspect of the NSF CPS proposal will be the construction of a multirotor flight platform based on theoretical results developed as part of this  project. This platform would provide new opportunities to test some CPS related issues involving computer-based control-implementation and communication-based decision realization, and would create a unique collaboration experience.  

  • Physical and Dynamical Meteorology (PDM) Program from NSF

This program accepts proposals related to the study of turbulent coherent structures using multi-rotor copters. These structures, on the order of a few hundred meters or more, have a large impact on the transport and exchange of heat, momentum, and mass, but cannot be sampled using fixed tower measurements. Multi-rotor copters with adaptive control schemes to be developed in this proposal allow the tracking and sampling of these structures in large detail. These measurements will help the development and evaluation of  new turbulence parameterization schemes for the improvement of numerical weather prediction models.