Coen C. de Visser

Controlled Flight Maneuvers of a Flapping Wing Micro Air Vehicle: a Step Towards the Delfly II Identification

The Delfly II Flapping Wing Micro Air Vehicle was flown in an external tracking chamber. It was possible to perform controlled flight-test maneuvers with an ornithopter that is capable of hover and forward flight, for system identification purposes. This was achieved by programming its autopilot to deflect the a control surface, while keeping the other surfaces at trimmed condition. Step, doublet and triplet inputs of 1/3, 2/3 and 4/3 of a second on the elevator, rudder and flapping frequency actuators were performed to excite the Delfly’s eigenmodes. These tests were carried out at different flight speeds, ranging from -0.5 to 8 m/s and with the ornithopter’s center of gravity at 83%, 74%, 44% and 42% of the wing root chord. As a result, it was possible to cover the Delfly’s flight envelope and collect data that will be used to build a dynamic and aerodynamic model of the Delfly. The selected inputs have shown to excite the Delfly in dampened oscillatory modes that can be compared to phugoid and short period for the longitudinal dynamics. The Delfly is highly affected by the rudder deflections. The results also reveal an unstable lateral mode similar to a spiral.

Black-box LTI modelling of flapping-wing micro aerial vehicle dynamics

This paper presents the development of black-box linear state-space models for the flight dynamics of a flapping-wing micro aerial vehicle (FWMAV), the DelFly. The models were obtained by means of system identification techniques applied to flight data recorded in a motion tracking chamber and describe the time-averaged dynamics of the vehicle in the proximity of specific stationary points in forward flight. Ordinary least squares and maximum likelihood-based estimation approaches were applied in the time domain, and decoupled models were identified for the longitudinal and the lateral dynamics. The availability of several different datasets additionally allowed for validation and for the estimation and comparison among each other of several separate models. Adequate models were obtained for both the longitudinal and the lateral dynamics. These reproduce the estimation data well and are also capable of predicting the response to validation inputs with a reasonable degree of accuracy, thus allowing for a simulation of the DelFly near the stationary points considered. The identified dynamics are stable and thus in agreement with the observed behaviour of the DelFly in the considered flight regime.

Selective Velocity Obstacle Method for Cooperative Autonomous Collision Avoidance System for Unmanned Aerial Vehicles

Autonomous collision avoidance system (ACAS) for Unmanned Aerial Vehicles (UAVs) is set as a tool to prove that they can achieve the equivalent level of safety, required in context of integrating UAVs flight into the National Airspace System (NAS). This paper focus on the cooperative avoidance part, aiming to define an algorithm that can provide avoidance between cooperative UAVs in general, while still be restricted by some common rules. The algorithm is named the Selective Velocity Obstacle (SVO) method, which is an extension of the Velocity Obstacle method. The algorithm gives guidelines for UAVs to select between three basic modes for avoidance, i.e., to Avoid, Maintain, or Restore. The variation of those three modes gives flexibility for UAVs to choose how will they avoid. By modeling the algorithm as a hybrid system, simulations on various UAVs encounters scenario were conducted and shows satisfying result. Monte Carlo simulations are then conducted to conclude the performance even more. Randomizing the initial parameters, including speed, attitude, positions and avoidance starting point, more than 10 encounter scenario were tested, involving up until five UAVs. A parameter called the Violation Probability is then derived, showing zero violations in the entire encounter samples.

Modeling a Flapping Wing MAV: Flight Path Reconstruction of the Delfly II

Flight identification techniques were used to excite the modes of a Flapping Wing Micro Air Vehicle the Delfly II. The flapper was flown in a flight chamber, where the position of eight markers placed on its structure was captured using high fidelity external tracking system, at 200Hz. The flight tests comprised step, doublet and triplet inputs on the control surfaces that were commanded by the autopilot during trimmed steady flight. The first step of the Two Step Method, known as Flight Path Reconstruction, was used to reconstruct the inputs and states, after assessing the quality of the recorded data, by means of differentiation and filtering. It was possible to identify two oscillatory modes: one longitudinal, similar to a phugoid, with a period of 1 second and a coupled lateral directional mode with a period of 0.9 seconds. The general aircraft equations of motion were used to estimate the aerodynamic forces and moments that act on the flapper during the maneuvers, under a set of non-flapping rigid body assumptions. The results point to coherent estimation of the aerodynamic forces. The moments around the lateral (y) and downwards (z) body axes, M and N moments respectively, seem to be well estimated for both the longitudinal and lateral/directional inputs on the elevator and rudder. However, the L moment around the longitudinal (x) body axis seems to have missing information around the maneuvers, due to the reconstruction method of the roll angles and roll rates. The good results point to the possibility of applying the second step of the Two Step Method for linear and non-linear aerodynamic model identification.

Velocity Obstacle Method for Non-cooperative Autonomous Collision Avoidance System for UAVs

Unmanned Aerial Vehicles (UAVs) are required to have Autonomous Collision Avoidance System (ACAS) to resolve conflicts, especially when flying in the National Airspace System (NAS). This paper focused on the non-cooperative concept for avoidance, where UAVs face rogue obstacle that does not cooperatively share their flight data, nor they follows the rule of the air. UAVswill rely on its on-board sensor for avoidance, in space and time range that is limited. The limitation make the entire process of sense, detect, and avoid (SDA) mostly not applicable, and sense and avoid (SA) manner is more preferable and safe. A method called SA-VO that extend the use of the known Velocity Obstacle (VO-) method is introduce to handle the problem. The method produce more efficient avoidance in the non-cooperative space than SA manner of avoidance, without fully run the SDA process. Probability of collision map based on ranges of obstacle range of positions and velocity is predefine to replace the detection part of SDA. The method is then tested using simulations of encounters between UAV and an obstacle. The simulations show that SA-VO can be a middle ground between the SDA and SA manner of avoidance.

SHERPA: a safe exploration algorithm for Reinforcement Learning controllers

The problem of an agent exploring an unknown environment under limited prediction capabilities is considered in the scope of using a reinforcement learning controller. We show how this problem can be handled by the Safety Handling Exploration with Risk Perception Algorithm (SHERPA) that relies on interval estimation of the dynamics of the agent during the exploration phase along with limited capability from the agent to perceive the presence of incoming fatal instances. An application to a simple quadrotor model is included to show the algorithm performance.

A Joint Sensor Based Backstepping Approach For Fault-Tolerant Flight Control of a Large Civil Aircraft

The sensor based backstepping (SBB) control law, based on singular perturbation theory and Tikhonov’s theory, is a novel incremental type high gain control approach. This Lyapunov function based method is not susceptible to model uncertainties since it uses measurements instead of reconstructed modeling variables. Considering these merits, we extended the SBB method, in this paper, to handle sudden structural changes in the fault tolerant flight control of a fixed wing aircraft. A new double-loop joint SBB attitude controller has been developed for a Boeing 747-200 aircraft using the backstepping technique. Compared with the double-loop NDI attitude control approach, the double-loop SBB attitude control setup enables the verification of the system stability and allows relatively more interaction between the angular rate loop and the angular loop. The benchmark rudder runaway and engine separation failure scenarios are employed to evaluate the proposed method. The simulation results show that the proposed joint SBB attitude control method can lead to a zero tracking error performance in the nominal condition and can guarantee the stability of the closed-loop system under the failures.

Aircraft Damage Identification and Classification for Database-driven Online Safe Flight Envelope Prediction

Safe flight-envelope prediction is essential for preventing aircraft loss of control after the occurrence of sudden structural damage and aerodynamic failures. Considering the unpredictable nature of such failures, many challenges remain in the process of implementing such a prediction system. In this paper, an approach to online safe flight-envelope prediction is proposed that is based on the retrieval of information from offline-assembled databases. One of the key steps of this approach is determining the structural damage of the state of the aircraft by using the identification, detection, and classification methods presented in this paper. The estimated damage cases will lead to structural damage indices in the database corresponding to those safe flight envelopes that are “closest” to the actual safe flight envelope of the damaged aircraft. The feasibility of the proposed database-driven approach is proved by simulation results, where three damage cases are successfully detected and classified.

A Framework for Calibrating Angular Accelerometers using a Motion Simulator

The angular accelerometer (AA) is a relatively new type of inertial sensor which is expected to play an important role in upcoming fault tolerant flight control systems. Calibrating such sensors is not trivial because specialized calibration equipment in the form of angular acceleration motion simulators are not widely available. In this research a highgrade commercially available angular position based motion simulators is used to generate both a real acceleration motion to excite the AA as well as an accurate acceleration reference to which the measurements from the AA can be compared for calibration. Through a careful analysis of the mechanical and electronic disturbance sources as well as the dynamic properties of the motion controller a sequence of low-pass (pre) filters is designed that filter out unwanted frequency content while leaving the true angular acceleration signal intact. The filters, together with the differentiators and calibration model estimators form a novel framework that is general enough to be used in any inertial sensor calibration setup involving position based motion simulators.

Reinforcement Learning Applied to a Quadrotor Guidance Law in Autonomous Flight

Autonomous flight of Unmanned Aerial Vehicles (UAVs) in unknown or uncertain environments can benefit from control methods that are able to learn and adapt to these conditions. This paper presents the setup and results of a high level reinforcement learning problem for both simulation and real flight tests. The problem provided is that of a quadrotor taking pictures of a disaster site. The environment is completely unknown at first and the agent must learn where the sites of interest are and the most efficient way to get there. The results show that the quadrotor agent can learn a converged, near optimal value function after many iterations. However, a non-converged value function can result in the same desirable actions with much fewer iterations. Furthermore in this paper, a research and test laboratory for ground robots and aerial vehicles is presented.