RPAS Navigation and Guidance Systems based on

Fourth Australasian Unmanned Systems Conference, 2014

RPAS Navigation and Guidance Systems based on GNSS and other Low-Cost Sensors Roberto Sabatini, Subramanian Ramasamy, Francesco Cappello, Alessandro Gardi School of Aerospace, Mechanical and Manufacturing Engineering RMIT University, Melbourne, VIC 3000, Australia [email protected] Abstract: A new navigation system for small size Remotely Piloted Aircraft Systems (RPAS) is presented, which is based on Global Navigation Satellite System (GNSS), Micro-Electro-Mechanical System (MEMS) based Inertial Measurement Unit (IMU), Vision Based Navigation (VBN) and other low-cost avionics sensors. The objective of this research is to design a compact, lightweight and relatively inexpensive Navigation and Guidance System (NGS) capable of providing the required navigation performance in all phases of flight of a small RPAS, with a special focus on precision approach and landing, where VBN techniques can be fully exploited in a multisensory integrated architecture. Additionally, the potential of carrier-phase GNSS for Attitude Determination (GAD) is explored and a six-degree-of-freedom (6-DoF) Aircraft Dynamics Model (ADM) is adopted to compensate for the MEMS-IMU sensor shortcomings in high-dynamics attitude determination tasks. The NGS data fusion architectures investigated are based on Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) approaches. After introducing the key hardware and software features of the NGS, the system performance is evaluated in a small RPAS integration scheme (i.e., AEROSONDE RPAS platform) by exploring a representative cross-section of this RPAS operational flight envelope, including a variety of high dynamics maneuvers and CAT-I to CAT-III precision approach tasks. The performance evaluation shows that the position and attitude accuracies of the proposed integrated navigation and guidance systems are compatible with the Required Navigation Performance (RNP) specified in the various RPAS flight phases, including precision approach down to CAT-II.

Introduction Although several research efforts have addressed the challenges involved with integration of Remotely Piloted Aircraft Systems (RPAS) into commercial airspace, there is still a substantial need to properly articulate the levels of airworthiness and operational capabilities required to enable RPAS to safely operate in non-segregated airspace [1]. RPAS are adopted for their ability to perform tasks with higher manoeuvrability and longer endurance than manned aircraft. In particular, small and medium-size RPAS hold good promise in carrying out a number of complex mission- and safety-critical tasks and are now an important part of the global aerospace and aviation market. According to recent studies, RPAS pose less risk to life and inanimate objects than their manned counterparts [2]. The distinctive abilities of RPAS can be attributed to the implementation of a number of enabling technologies including avionics systems miniaturization [3], evolving information and communication systems, and the widespread availability of a number of high-performance sensors with steadily decreasing cost, weight and volume. To accomplish a number of diverse missions in the foreseeable operational scenarios, RPAS will need enhanced navigational capabilities that fulfil the Required Navigational Performance (RNP) and Reduced Vertical Separation Minima (RVSM) expected of manned aircraft [1]. In order to maintain an adequate separation between manned aircraft and RPAS in cooperative and non-cooperative scenarios, the Performance Based Operations (PBO) approach defines a set of Communication, Navigation and

Surveillance (CNS) performance requirements related to the different flight phases [4]. In particular, RPAS specific navigation requirements comprise: physical characteristics of the sensors including size, weight and volume, support requirements such as electrical power, accuracy and precision, and meeting the prescribed system accuracy, integrity, availability and continuity. The primary navigation sensors that are currently employed in most RPAS are based on satellite navigation and inertial systems [5]. Global Navigation Satellite System (GNSS) sensors provide high-accuracy position and velocity data using pseudorange, carrier phase, Doppler observables or various combinations of these measurements. A number of global and regional navigation satellite systems including Global Positioning System (GPS), Galileo, GLObalnaya NAvigatsionnaya Sputnikovaya Sistema (GLONASS), BeiDou (Compass), Indian Regional Navigational Satellite System (IRNSS) and QuasiZenith Satellite System (QZSS) are currently in service or being planned to provide accurate positioning services. In this research, GPS Standard Positioning Service (SPS) pseudorange measurements are considered for position and velocity computations. Due to the low volume/weight and cost of current carrier-phase GNSS receivers, and the high accuracy attainable, interferometric GNSS technology is currently an excellent candidate for RPAS applications [6-8]. Inertial Measurement Unit (IMU) data are integrated over time to obtain position, velocity and attitude estimates in an Inertial Navigation System (INS). Low-cost and low-weight Micro-Electro-Mechanical

Fourth Australasian Unmanned Systems Conference, 2014

System (MEMS) based IMU are particularly well suited for small size RPAS applications. Since the MEMS-based INS measurements are not very accurate and drift rapidly, they are typically augmented with GNSS and other data. Additionally, the recent advances in electro-optics technology have made Vision Based Navigation (VBN) a viable option for a number of mission- and safety-critical applications [9]. VBN techniques provide passive and cost effective solutions and are not subject to the same limitations of other sensors. In particular, the VBN sensors are self-contained and autonomous, which enables them to be used as an alternative to more traditional sensors and systems (e.g., INS, GNSS and integrated GNSS/INS), especially in precision approach and landing applications, which are the most demanding and safety-critical flight phases [10, 11, 12]. VBN approaches are categorised under Simultaneous Localisation and Mapping (SLAM), visual odometry and terrain-aided/mapaided methods [10]. The increase in precision and accuracy of vision based approach/landing can be attributed to the shift in control technology from basic nonlinear types to intelligent, hybrid and robust control [13]. Vision based sensors can be also employed in GNSS-denied environments including jamming and spoofing conditions [14]. Hyper spectral images have been also used on a multi-rotor Hyper spectral Unmanned Aircraft System (HyperUAS) carrying an on-board spectroradiometer coupled with a dual frequency GPS and an inertial sensor [15]. A challenging task is to provide autonomous landing of RPAS on aircraft carriers, since a number of factors including low cost sensors, unknown movements of the landing surface and external disturbances make it hard to generate accurate relative attitude estimation, which is sufficient for landing. Architectures that overcome these limitations even in the presence of wind disturbances have been proposed in [16]. Manoeuvre-based navigation methods can aid small RPAS in urban operations involving high levels of autonomy for tasks in a cluttered environment [17]. Expanding the spectrum of sensors that can be adopted, Aircraft Dynamics Models (ADM) have been proposed recently to compensate for the VBN and MEMS-IMU sensor shortcomings [18]. The ADM virtual sensor is essentially a KnowledgeBased Module (KBM), which is used to augment the navigation state vector by predicting the RPAS flight dynamics (aircraft trajectory and attitude motion). The ADM employs either a three-degree-of-freedom (3-DoF) or a six-degree-of-freedom (6-DoF) variable mass model with suitable controls and constraints applied in different phases of flight. Generally, the image processing frontend is susceptible to false detection of the horizon if other high-contrast edges are present in the image [18, 19]. Therefore, an Extended Kalman Filter (EKF) is typically

implemented to filter out the spurious results. The ADM is also used to compensate for the MEMSIMU sensor shortcomings experienced in highdynamics attitude tasks and, in this case, the EKF can be replaced by an UKF [19]. Additionally, the UKF can be also used to pre-process the ADM navigation data. The pre-filtering of the ADM virtual sensor measurements aids in achieving a reduction of the overall position/attitude error budget and, most importantly, a considerable reduction in the ADM reinitialisation time (i.e., increased validity time of the ADM data). In line with the above discussions, the main objective of our research is to develop a lowcost and low-weight/volume Navigation and Guidance System (NGS) based on GNSS and other low-cost and low-weight/volume sensors, capable of providing the required level of performance in all flight phases of a small to medium size RPAS. In addition to developing integrated navigation sensors, the data provided by the NGS are used to optimise the design of a hybrid control system, tailored for VBN, which employs Fuzzy logic and Proportional/ Integral/ Derivative (PID) techniques.

Sensor Choices for Data Fusion Recent research activities [4, 5, 11, 12, 18, 19] addressed various sensors and architectures for multi-sensor data fusion. VBN techniques use optical sensors (visual or infrared cameras) to extract features from images of the surrounding environment, which are then used for localization. A number of effective algorithms for VBN processing have been developed and used for RPAS applications. In order to achieve better results, attitude can be estimated using different techniques such as sky/ground segmentation and color-based separation methods [18]. The main vision based methods for navigation include Simultaneous Localization and Mapping (SLAM) [20], terrainbased navigation and navigation based on feature recognition [21]. RPAS vision based systems have been developed for various applications ranging from autonomous landing to obstacle avoidance. Though several VBN sensors and techniques have been developed to date, the vast majority of VBN processing schemes falls into one of the following two categories [22]: Model-based Approach (MBA) and Appearance-based Approach (ABA). MBA uses feature tracking in images and creates a ThreeDimensional (3D) model of the workspace in which the RPAS operates. MBA has been extensively investigated in the past and is the most common technique currently implemented for VBN. The ABA approach has a disadvantage that it requires a large amount of memory to store images and is computationally more costly than MBA. However, due to improvements in computer technology, this technique has become a viable solution for many applications. In this research, the ABA approach is

Fourth Australasian Unmanned Systems Conference, 2014

adopted for the design of the VBN sensor system. The VBN process is depicted in Figure 1. The key frames represent the visual route that the RPAS is required to follow. The figure shows that the key frame 2 is identified as the starting point of the visual route during the localisation process. Key frame 1 Key frame 2 (start of visual route) Key frame 3 Key frame 4 Key frame 5 (end of visual route)

Figure 1 Conceptual representation of the VBN process on final approach.

The Image Processing Module (IPM) of the VBN system detects horizon and runway centreline from the images and computes the aircraft attitude, body rates and deviation from the runway centreline. The functional architecture of the IPM is illustrated in Figure 2. Current view from on-board camera

Current key image frame from memory

Extraction and computation of horizon using Canny edge detector

Computation of horizon using Canny edge detector

Extraction and computation of runway centreline location using Hough transform

Computation of runway centreline location using Hough transform

(2)

Image comparison and computation of pitch difference, roll difference and deviation from runway centreline

Computation of optical flow for all points on the detected horizon

UA attitude

Pitch difference

Body rates

Roll difference

(carrier-phase) has been also considered in recent years, mostly for spacecraft applications (i.e., replacing or aiding traditional sun-sensors, horizontrackers, star-trackers, magnetometers, etc.), and for manned aircraft and maritime applications [6, 7]. Due to the low volume/weight of current carrierphase GNSS receivers, and the very high accuracy attainable notwithstanding their lower cost, interferometric GNSS technology is becoming a good candidate for RPAS applications as well. Various computational methods have been developed in the past for GAD systems. Independently from the method selected, since GAD errors are heavily affected by the length of the baselines used (longer baselines giving smaller errors), some efficient geometric algorithms have been proposed for baseline selection in the presence of redundant satellite measurements. The accuracy of GAD systems is affected by several factors including the selected equipment/algorithms and the specific platform installation geometry, with the baseline length and multipath errors being the key elements dominating GAD systems performance [7]. The phase measurement of the GNSS signal carrier allows determining the relative displacement of the antennae in the body reference frame. This information is directly related to the attitude of the vehicle. The carrier phase measurement is given by: (1) where is integer ambiguity, is the clock bias, represents the atmospheric delays, and represents the noise and multipath errors. The displacement of the antenna baseline (b) with respect to the Line-ofSight (LOS) of the GNSS signal is given by [7]:

Deviation from runway centreline

Figure 2 Functional architecture of the IPM

Input data for the ADM are made available from aircraft physical sensors (i.e., aircraft data network stream) and from ad-hoc databases [19]. Interferometric GNSS for Attitude Determination (GAD) is adopted for augmenting attitude information. GNSS Carrier Phase Measurements (CFM) are utilised for attitude estimation. The concept of replacing/augmenting traditional attitude sensors with GNSS interferometric processing

where the phase difference ∆/360 is proportional to the projection of the baseline (b) on the LOS, is the wavelength and a proportionality constant. Since the antennae are placed at different locations, the phase measurements of the incoming GNSS signal carrier are different for each antenna. By knowing the integer number of cycles travelled by the carrier (N), it is possible to determine the vehicle attitude. Figure 3 illustrates the GNSS attitude determination process. The differencing measurement from antenna 1 and 2 is utilized to resolve ambiguity in a constrained manner. Based on the baseline length value obtained, the attitude is determined. Antenna 1

Differencing measurement

Constrained ambiguity resolution

Baseline length

Antenna 2

Figure 3 GNSS attitude determination

Attitude determination

Fourth Australasian Unmanned Systems Conference, 2014

When using GNSS for attitude determination it is sufficient that only two satellites are in view due to the following considerations: 

Common time reference Measurements are independent from the error at the receiver clock as this is the same for the measurements performed by each antenna.



Baseline setting The relative position of the antennae on the vehicle is known a priori; this eliminates another unknown factor, which reduces the number of satellites required.

Based on recent research, signals of opportunity can also be adopted. These are existing radio frequency (RF) signals in the airspace surrounding the RPAS, which tend to have much higher power levels and wider coverage especially in urban environments than GNSS signals. For instance, information from WiFi, cellular base stations, radio communication signals and others can be used in the data fusion process. However, in the following, we will concentrate on GNSS, MEMS-based INS, VBN and ADM for the design of a low-cost and lowweight/volume NGS for RPAS.

Loose, tight and deep integration approaches can be employed for multi-sensor data fusion at navigation solution, measurements and signal processing levels respectively. A loosely coupled integration method supports the integration of Commercial Off-TheShelf (COTS) and low-cost navigation sensors [24]. Therefore, our NGS employs a loosely coupled approach with sensor pre-selection (Boolean decision logics) and centralised data fusion based on Extended/Unscented Kalman Filters (EKF/UKF) algorithms. The multi-sensor data fusion process is illustrated in Figure 4. PILOT INTERFACE Navigation and guidance mode switch

Global Navigation Satellite System

VIGA (Degraded mode of EVIGA)

Inertial Measurement Unit

UVIGA

Vision-Based Sensor

Aircraft Dynamics Model

Sensors Sensor processing and data sorting Multi sensor data fusion techniques Extended Kalman Filter

Linear Systems Analytical Linearization Original KF

Unscented Kalman Filter

Position, Velocity and Attitude (PVA) best estimates

Figure 4 Multi-sensor data fusion process

Kalman Filters Nonlinear Systems Stochastic Linearization

Extended KF Iterated KF

Multi-Sensor Data Fusion

VIG (Degraded mode of VIGA)

Position, velocity and attitude measurements are obtained from GNSS. MEMS-based INS provides position and velocity data while attitude measurements are also obtained both from INS and VBN sensors. ADM, acting as a virtual sensor, also provides attitude measurements. EKF and UKF are used for multi-sensor data fusion. The measurements from VBN sensors, GNSS and MEMS-IMU are employed in the VBN/IMU/GNSS (VIG) NGS architecture. By including measurements from ADM, the VBN/IMU/GNSS/ADM (VIGA) system architecture is realized. EKF is used in both VIG and VIGA architectures. In UVIGA architecture, the EKF is replaced by an UKF. A classification of the estimation methods based on Kalman Filter (KF) is illustrated in Figure 5. The original KF requires that the system measurements remain linearly related to the state vector. In most practical applications this condition is not verified and, therefore, analytical and stochastic linearization methods are employed.

SPKF

SSUKF

UKF

Sq. Root UKF

Additive UKF

Other SPKF

Figure 5 Classification of Kalman filters (Adapted from [23])

The Extended Kalman Filter (EKF) extends the scope of the original KF to nonlinear optimal filtering problems by forming a Gaussian approximation to the state and measurement vector distribution using a Taylor series based transformation. The EKF accounts for nonlinearities by linearizing the system about its last-known best estimate with the assumption that the error incurred by neglecting the higher-order terms is small in comparison to the first-order terms. The drawback in adopting an EKF is the complexity involved in the derivation of the Jacobian matrices and the linear approximations of the nonlinear functions. Furthermore, the accuracy of propagated mean and covariance is limited to first order due to the truncation in the linearization process. Recent studies have suggested that implementing the EKF gives rise to a number of performance flaws, where most deficiencies can be addressed by the Unscented Kalman Filter (UKF). Sigma-point Kalman Filters (SPKFs), such as the UKF, provide derivative-free higher-order approximations by fitting a Gaussian distribution rather than approximating an arbitrary nonlinear function as the EKF does. For navigation applications, the UKF is more robust and accurate than EKF and provides better convergence characteristics. The additive UKF is used to reduce the number of mathematical calculations performed

Fourth Australasian Unmanned Systems Conference, 2014

for each iteration and also reduces the computational load of the filter [23]. The square-root UKF is adopted to prevent numerical instabilities and to reduce the computational cost. The Spherical Simplex UKF (SSUKF) is employed in cases where a minimum set of sigma points is required [23]. A Particle Filter (PF) is a very flexible approach that allows an implementation of a recursive Bayesian filter through Monte Carlo simulations. It can be applied to a wide spectrum of dynamic state-space models covering linear, non-linear, Gaussian, nonGaussian, stationary, non-stationary, continuous, discrete and hybrid models. In our research, it was observed that the image processing frontend was susceptible to false detection of the horizon if any other strong edges were present in the images; an EKF is implemented to filter out these incorrect results to provide best estimates. Furthermore, an UKF is implemented to increase the accuracy of the measurements and it also aids in pre-filtering the ADM measurements. It is assumed that the motion model of the aircraft is disturbed by uncorrelated zero-mean Gaussian noise. The state vector consists of the roll angle, pitch angle and body rates of the aircraft. The EKF measurement model is defined as: (3) where is the measurement vector, is the design matrix, is the state vector, is the measurement noise and k is the kth epoch of time, . The successive states are given by: (4)

the current measurement and the measurement and can be described as:

predicted (8)

The Kalman gain is used to quantify the influence of new information present in the innovation vector on the estimation of the state vector and can be considered as a weight factor. It is equal to the ratio of the uncertainty on the current measurement and the uncertainty on the predicted one. This gain is given by: (9) where is the measurement noise covariance matrix. For the process model defined here, the state vector of the system composed of error in position, velocity, and attitude, is given by: (10) The covariance matrix of the model is given by:

(11)

The discrete process noise matrix is defined as follows:

(6)

(12) The UKF is implemented primarily to increase the ADM validity time. The UKF is a recursive estimator and is based on unscented transformations in which unscented transforms are used for calculating the statistics of a random variable that goes through a nonlinear transformation. The process model of the UKF is based upon a set of sigma points. The sigma points, are selected based on the mean and covariance of . The sigma points are obtained by: (13) (14) where P is the lower triangular matrix of the Cholesky factorisation, S and is the control parameter of the dispersion distance from the mean estimate in the computation of the sigma point matrix, After the sigma points are calculated, update of time is performed for each time step and is given by: (15)

(7)

(16)

where is the Kalman gain matrix at epoch, k+1 and is the innovation vector at epoch, k+1. The innovation vector represents the difference between

(17)

where is the state vector at epoch k+1, is the state transition matrix from epoch k to k+1, is the shaping matrix and is the process noise. The EKF comprises of prediction and update steps. The prediction algorithm estimates the state vector and computes the corresponding covariance matrix from the current epoch to the next one using the state transition matrix characterizing the process model described by: (5) where represents a predicted value computed by the prediction equations and refers to updated values obtained after the correction equations. The process noise at a certain epoch k is characterized by a covariance matrix, . The updating equations correct the predicted state vector and the corresponding covariance matrix using the measurement model as follows:

Fourth Australasian Unmanned Systems Conference, 2014

The measurement update equations are: (18) (19) (20) (21) where is the nonlinear function used for propagation of the sigma points, is the component of the vector is an a-priori estimate of conditioned by all prior measurements except the one at time , is an a-posteriori estimate of , conditioned by measurements up to time .

Integrated NGS Architectures Four different integrated navigation system architectures are identified, including EKF based VBN-IMU-GNSS (VIG), VBN-IMU-GNSS-ADM (VIGA), VBN-IMU-GNSS-GAD (VIGGA) and UKF based VIGA (UVIGA) system. The VIG architecture uses VBN at 20 Hz and GPS at 1 Hz to augment the MEMS-IMU running at 100 Hz. The VIGA architecture includes the ADM (computations performed at 100 Hz) to provide attitude channel augmentation while the VIGGA architecture includes GNSS to provide attitude channel augmentation. The navigation system outputs are fed to a hybrid Fuzzylogic/PID controller designed for the AEROSONDE RPAS and capable of operating with stand-alone VBN, as well as with other sensors data. The VIG architecture is illustrated in Figure 6. The INS provides measurements from gyroscopes and accelerometers which are fed to a navigation processor. GNSS provides raw pseudorange measurements which are processed by a filter to obtain position and velocity data [25]. A similar process is also applied to the INS and VBN attitude angles, whose differences are incorporated in the EKF measurement vector. The EKF provides estimates of the Position, Velocity and Attitude (PVA) errors, which are then removed from the sensor measurements to obtain the corrected PVA states. The corrected PVA and estimates of accelerometer and gyroscope biases are also used to update the INS raw measurements. The data sorting algorithm is based on Boolean decision logics and allows automatic selection of the sensor data based on pre-defined priority criteria. The sorted data is then fed to the estimator to obtain the best estimate values. The VIGA architecture is illustrated in Figure 6 (additional ADM block and interfaces are highlighted). As before, the INS position and velocity provided by the navigation processor are compared with GNSS data to form the measurement input of EKF. Additionally, in this case, the attitude data provided by the ADM and the INS are compared to feed the EKF at 100 Hz, and the attitude

data provided by the vision-based sensors and INS are compared at 20 Hz and input to the EKF. As in the VIG architecture, the EKF provides estimations of PVA errors, which are removed from the INS measurements to obtain the corrected PVA states. The attitude best estimate is compared with the INS attitude to obtain the corrected attitude. During the landing phase, the attitude best estimate is compared with the VBS attitude to obtain the corrected attitude. The GNSS attitude determination is integrated to the VIG Navigation System to form the VIGGA architecture as illustrated in Figure 7. In addition to the VIG architecture, the raw carrier-phase measurement is used for GAD system (GADS) to provide GADS attitude. The attitudes from INS, VBS and GADS form the attitude measurement inputs for the data fusion block. In this case also, the corrected PVA and estimates of accelerometer and gyroscope biases are used to update INS raw measurements. In the UVIGA architecture, EKF is replaced by an UKF (Figure 8). Additionally, an UKF is also adopted to pre-process the ADM navigation solution. The ADM operates differently to that of the VIGA system running in parallel to the centralised UKF. The pre-filtering of the ADM virtual sensor measurements aids in achieving reduction of the overall position and attitude error budget and significantly reduces the ADM reinitialisation time. PVA measurements are obtained as state vectors from both the centralised UKF and ADM/UKF. These measurements are then fed into an error analysis module in which the measurement values of the two UKFs are compared. The error analysis block includes the primary sensors (GNSS, MEMS-IMU and VBN) and it is used to compare the error values with the ADM error to obtain the corrected PVA states.

Simulation Case Study The AEROSONDE model from Unmanned Dynamics LLC is used for modelling and simulation. The AEROSONDE RPAS is a small autonomous aircraft used in weather-reconnaissance and remotesensing missions. This model is part of the AeroSim Blockset implemented in MATLAB/Simulink. The AeroSim Blockset provides components for rapid development of non-linear 6-DoF dynamic models. The library also includes Earth models (geoid references, gravity and magnetic fields) and atmospheric models. The AEROSONDE RPAS model can be interfaced with simulators such as FlightGear and Microsoft Flight Simulator to allow visualisation of the aircraft trajectory. The inputs to the AEROSONDE model include control surface deflections in radians, throttle input, mixture and ignition. The model outputs the various aircraft states such as the position in the Earth-fixed frame, velocity, attitude and attitude rates.

Fourth Australasian Unmanned Systems Conference, 2014

Sensors

Inertial Measurement Unit Global Navigation Satellite System

Sensor Processing and Data Sorting Corrected Position and Velocity

Accelerometers Gyroscopes Measurements

Position Velocity Best Estimate

GNSS Position IMU Position GNSS Velocity

GNSS Filter Raw Pseudorange Measurement

Data fusion EKF

IMU Velocity

Attitude Best Estimate

IMU Attitude ADM Attitude Corrected Attitude IMU Attitude VBN Attitude

Vision-Based Navigation Processor

VBN Attitude

Corrected Position and Velocity

+

IMU Position and Velocity Navigation Processor

Aircraft Dynamics Model

Vision-Based Sensor

Multi-Sensor Data Fusion

-

Corrected Attitude

+

VBN Attitude

IMU Attitude

Figure 6 VIG and VIGA architectures Sensors

Sensor Processing and Data Sorting

Multi-sensor Data Fusion

Corrected Position and Velocity Inertial Navigation System Global Navigation Satellite System

GADS

+

INS Position and Velocity Accelerometers Gyroscopes Measurements

Navigation Processor

Position and Velocity Best Estimate

GNSS Filter Raw Pseudorange Measurement

Data fusion EKF

GADS Attitude Attitude Best Estimate

Aircraft Dynamics Model

Vision-Based Sensor

Corrected Position and Velocity

-

Vision-Based Navigation Processor

VBN Attitude

Corrected Attitude

+

Figure 7 VIGGA architecture Sensors

Sensors Processing and Data Sorting

Multi-sensor Data Fusion

Error Analysis

Corrected Position and Velocity Inertial Navigation Processor

Inertial Navigation Sensors

Accelerometers Gyroscopes Measurements

Global Navigation Satellite System

GNSS Navigation Processor Raw Pseudorange Measurements

Aircraft Dynamics Model

UKF/Attitude Processor

+

INS Position and Velocity

- Corrected VIG Position and Velocity

GNSS Position INS Position

Position and Velocity Best Estimate

Error Analysis

GNSS Velocity INS Velocity INS Attitude

VIGA+ Position and Velocity

Data fusion UKF Attitude Best Estimate

ADM Position and Velocity

ADM Attitude

Error Analysis

Corrected Attitude INS Attitude Vision-Based Sensor

Vision-Based Navigation Processor

Raw VBN Measurements

Figure 8 UVIGA architecture

-

VBN Attitude

+ INS Attitude VBN Attitude

Corrected VIG Attitude

VIGA+ Attitude

Fourth Australasian Unmanned Systems Conference, 2014

In order to perform the GNSS attitude determination for the AEROSONDE, 3-5 GNSS antennae were selected in order to optimize the length of the baselines. The position of the 3 antennae configuration in the AEROSONDE RPAS is illustrated in Figure 9.

The multi-sensor architectures were tested in an appropriate sequence of flight manoeuvres representative of the AEROSONDE RPAS operational flight envelope. The duration of the simulation is 700 seconds. The list of the different simulated flight manoeuvres and associated control inputs is provided in Table 1. The 3D trajectory plot of RPAS flight phases is shown in Figure 10.

Antenna 1

Table 1 Flight manoeuvres and control inputs Distance B

Flight manoeuvre Straight climb (take off) Left turning helical climb Straight and level Right loiter Right helix descent Straight descent

Antenna 2 Distance A

Distance C

Antenna 3

Figure 9 AEROSONDE RPAS antenna location

Required roll [°]

Required pitch [°]

Time [s]

0

10

50

-10

8

200

0

3.5

50

10

3.5

100

10

3

250

0

-2

50

Figure 10 3D trajectory plot of AEROSONDE RPAS flight phases

The ADM validity times for estimated position in east are shown in Figure 11. U-VIGA Estimated Position Error [East] 25

T=56.2 20 15

T=20

Error [m]

10

T=30

5 0 -5 -10 -15 -20 -25 0

ADM-UKF GPS-UKF CAT I CAT II CAT III 10

20

30

40

Time, T = [0,130] s

Figure 11 ADM validity times

50

60

70

80

Table 2 shows the position error statistics obtained for the four NGS architectures. Table 3 shows the attitude error statistics of the NGS architectures. During the initial VIGA simulation runs, it was evidenced that the ADM data cannot be used effectively without being reinitialised regularly. For the AEROSONDE RPAS manoeuvres listed in Table 1, it was found that an adequate period between ADM re-initialisation was in the order of 20 seconds. The VIG velocity error time histories show that GNSS is the dominating sensor for velocity computations but a significant improvement is obtained with the VIG system on the accuracy of the vertical data. Although the UVIGA provide comparable performance in terms of horizontal/vertical position and attitude data, a

Fourth Australasian Unmanned Systems Conference, 2014

Table 2 Position error statistics

significant improvement in the stability (i.e., solution validity time) of the ADM output data was observed when these data are pre-filtered (UKF). In particular, a comparison of ADM solutions in the UVIGA and VIGA architectures is presented in Table 4. The validity time before the solution exceeds the RNP 1 threshold in the climb phase is 78 sec and, in the final approach phase, the ADM solution exceeds the CAT I, CAT II and CAT III limits at 56 sec, 30 sec and 20 sec respectively (the VIGA is compliant with CAT I up to 35 sec, CAT II up to 20 sec and CAT III up to sec to 16 sec). The vertical channel satisfies CAT III and CAT II requirements up to approximately 100 sec and CAT I requirements up to 365 sec. Table 5 shows a comparison of the horizontal and vertical accuracy (RMS-95%) with the required accuracy levels for precision approach obtained from the NGS architectures as recommended by the International Civil Aviation Organization (ICAO) [26, 27]. The accuracy performances of the NGS systems are in line with Category Two (CAT II) precision approach requirements. In the UVIGA architecture, the ADM navigation solution is useful for an extended period of time and this behavior can be instrumental in aiding or replacing other sensors in case of temporary data losses or degradations.

North Position [m] μ σ 0.37 1.91 0.36 1.90 0.37 1.90 0.37 1.40

NGS System VIG VIGA VIGGA UVIGA

East Position [m] μ σ -0.49 1.94 -0.48 1.94 -0.48 1.94 -0.40 1.73

Down Position [m] μ σ 0.19 2.49 0.17 2.44 0.16 2.44 0.12 2.25

Table 3 Attitude error statistics NGS System VIG VIGA VIGGA UVIGA

Pitch ( ) [degrees] μ σ 0.013 0.048 0.005 0.040 0.004 0.038 0.005 0.040

Roll ( ) [degrees] μ σ -0.007 0.352 -0.006 0.313 -0.006 0.278 -0.005 0.219

Heading ( ) [degrees] μ σ -0.011 0.052 -0.011 0.044 -0.010 0.042 0.011 0.042

Table 4 ADM validity time ADM validity time [sec] VIGA UVIGA 64 78 35 56 20 30 16 20

Accuracy threshold RNP 1 CAT I CAT II CAT III

Table 5 NGS architectures precision approach position error statistics

Category of approach CAT I CAT II CAT III

Horizontal Accuracy (m) 2D RMS-95% Required VIG VIGA VIGGA 16 6.9 5.3 5.2 5.2 4.1

Conclusions The research activities performed to design a lowcost and low-weight/volume integrated NGS suitable for small size RPAS applications were described. Various candidate sensors were considered for integration in the NGS including VBN, GNSS and MEMS-IMUs, with and without augmentation from ADM. The possibility of using GAD outputs in the data fusion algorithms was also investigated. Four different architectures were defined and studied in detail. They were an EKF based VBN-IMU-GNSS (VIG), VBN-IMU-GNSS-ADM (VIGA), VBNIMU-GNSS-GAD-ADM (VIGGA) integrated systems and an UKF based UVIGA system. While the VIGA architecture uses unfiltered ADM data, the UVIGA employs an UKF for pre-filtering the ADM attitude solution, so to increase the ADM attitude solution stability. Simulation case studies of all architectures were performed on an AEROSONDE RPAS performing manoeuvres representative of the aircraft operational flight envelope. Compared to the

UVIGA 4.9

Required 4 2 2

Vertical Accuracy (m) RMS-95% Down VIG VIGA VIGGA 1.8

1.7

1.7

UVIGA 1.7

VIGA/VIGGA architectures, the UVIGA showed as significant improvement in the ADM solution stability time. Furthermore, the NGS horizontal/vertical position accuracy were in line with CAT-II precision approach requirements. Future research will address uncertainty analysis and possible synergies of the NGS architectures with GNSS space, ground and avionics based integrity augmentation systems [28-30], cooperative and noncooperative Detect-and-Avoid (DAA) functions [3134] and novel ATM and avionics systems [35-39].

References 1. Martin TL, Campbell DA, RPAS Integration within An Australian Atm System: What Equipment And Which Airspace, Unmanned Aircraft Systems (ICUAS), 2014 International Conference, pp. 656-668, 2014. 2. Melnyk R, Schrage D, Volovoi V, Jimenez H, A Third-Party Casualty Risk Model For Unmanned

Fourth Australasian Unmanned Systems Conference, 2014

Aircraft System Operations, Reliability Engineering & System Safety, Vol. 124, pp. 105116, 2014. 3. Logan MJ, Chu J, Motter MA, Carter DL, Ol M, Zeune C, Small UAV Research And Evolution In Long Endurance Electric Powered Vehicles, AIAA Infotech@ Aerospace 2007 Conference and Exhibit, AIAA Paper, pp. 7-10, 2007. 4. Degarmo MT, Issues Concerning Integration of Unmanned Aerial Vehicles In Civil Airspace, The MITRE Corporation - Center for Advanced Aviation System Development, 2004. 5. Wang J, Garratt M, Lambert A, Wang JJ, Han S, Sinclair D, Integration of GPS/INS/Vision Sensors to Navigate Unmanned Aerial Vehicles, IAPRSSIS, Vol. 37, pp. 963-9, 2008. 6. Sabatini R, Rodriguez L, Kaharkar A, Bartel C, Shaid T, Carrier-Phase GNSS Attitude Determination and Control System for Unmanned Aerial Vehicle Applications, ARPN Journal of Systems And Software, vol. 2, 2012. 7. Sabatini R, Rodríguez L, Kaharkar A, Bartel C, Shaid T, Zammit-Mangion D, Low-Cost Navigation And Guidance Systems for Unmanned Aerial Vehicles—Part 2: Attitude Determination and Control, Annual of Navigation, vol. 20, pp. 97-126, 2013. 8. Rhudy M, Gu Y, Napolitano MR, Low-Cost Loosely-Coupled Dual GPS/INS for Attitude Estimation with Application to a Small UAV, AIAA Guidance Navigation and Control Conference, Boston, MA, 2013. 9. Sanfourche JDM, Le Besnerais G, de Plinval H, Israel J, Cornic Ph, Treil A, Watanabe Y, Plyer A, Perception for UAV: Vision-Based Navigation and Environment Modeling, Aerospace Lab Journal, 2012. 10. Dumble SJ, Gibbens PW, Airborne Vision-Aided Navigation Using Road Intersection Features, Journal of Intelligent & Robotic Systems, pp. 120, 2014. 11. Sabatini R, Richardson MA, Bartel C, Shaid T, Ramasamy S, A Low-cost Vision Based Navigation System for Small Size Unmanned Aerial Vehicle Applications, Journal of Aeronautics and Aerospace Engineering, vol. 2, no. 3, May 2013. DOI: 10.4172/21689792.1000110 12. Sabatini R, Bartel C, Kaharkar A, Shaid T, Rodriguez L, Zammit-Mangion D, et al., Lowcost navigation and guidance systems for Unmanned Aerial Vehicles. Part 1: Vision-based and Integrated Sensors, Annual of Navigation, 2012. 13. Gautam A, Sujit PB, Saripalli S, A survey of autonomous landing techniques for UAVs,

International Conference on Unmanned Aircraft Systems (ICUAS), pp. 1210-1218, 2014. 14. Pestana J, Sanchez-Lopez JL, Saripalli S, Campoy P, Computer Vision based General Object following for GPS-denied Multirotor Unmanned Vehicles, American Control Conference (ACC), pp. 1886-1891, 2014. 15. Lucieer A, Malenovský Z, Veness T, Wallace L, HyperUAS—Imaging Spectroscopy from a Multirotor Unmanned Aircraft System, Journal of Field Robotics, vol. 31, pp. 571-590, 2014. 16. Ling K, Chow D, Das A, Waslander SL, Autonomous Maritime Landings for Low-Cost VTOL Aerial Vehicles, 2014 Canadian Conference Computer and Robot Vision (CRV), pp. 32-39, 2014. 17. Myung H, Kanade T, Maneuver-based autonomous navigation of a small fixed-wing UAV, 2013 IEEE International Conference on Robotics and Automation (ICRA), pp. 3961-3968, 2013. 18. Sabatini R, Cappello F, Ramasamy S, Gardi A, Clothier R, An Innovative Navigation and Guidance System for Small Unmanned Aircraft using Low-Cost Sensors, Journal of Aircraft Engineering and Aerospace Technology, 2015. (In press) 19. Sabatini R, Ramasamy S, Gardi A, Rodriguez L, Low-cost Sensors Data Fusion for Small Size Unmanned Aerial Vehicles Navigation and Guidance, International Journal of Unmanned Systems Engineering, vol. 1, no. 3, pp. 16-47. 2013. doi: 10.14323/ijuseng.2013.11 20. Kim J, Sukkarieh S, 6 DoF SLAM aided GNSS/INS navigation in GNSS denied and unknown environment, Journal of Global Positioning Systems, vol. 4, pp.120-128, 2005. 21. Kais M, Dauvillier S, Fortellea A, Masakid I, Laugier C, Towards Outdoor Localization using GIS, Vision System and Stochastic Error Propagation, International Conference on Autonomous Robots and Agents, 13th-15th December, Palmerston North, New Zealand, 2004. 22. Desouza GN, Kak AC, Vision for Mobile Robot Navigation: a Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 2, pp. 237 – 267, 2002. 23. Castrejon-Lozano JG, Dzul A, ThreeDimensional Localization, Unmanned Aerial Vehicles – Embedded Control, John Wiley & Sons, Inc., 2010. 24. Sabatini R, Development of a New MIDS Relative Navigation Kalman Filter and Suitable Interfaces (MIL-STD-1553B and Ethernet) for Slow-moving Platforms Equipped with INS,

Fourth Australasian Unmanned Systems Conference, 2014

GPS, GPS/INS and Dead-Reckoning Navigation Sensors, MIDS International Program Office Seminar, San Diego, CA (USA), April 2009. 25. Sabatini R, Rodríguez L, Kaharkar A, Bartel C, Shaid T, Satellite Navigation Data Processing for Attitude Determination and Control of Unmanned Air Vehicles, European Navigation Conference 2012 (ENC 2012), Gdansk (Poland), April 2012. 26. CAA Safety Regulation Group, GPS integrity and potential impact on aviation safety, Paper 2003/09, Civil Aviation Authority, Cheltenham, UK, 2004. 27. ICAO, Aeronautical Telecommunications Volume 1: Radio Navigation Aids. (6th ed.), Annex 10 to the Convention on International Civil Aviation, International Civil Aviation Organization, Montreal, Quebec, Canada, 2006. 28. Sabatini R, Moore T, Hill C, A New Avionics Based GNSS Integrity Augmentation System: Part 1 – Fundamentals, Journal of Navigation (Royal Institute of Navigation), vol. 66, no. 3, pp. 363-383, May 2013. DOI: 10.1017/S0373463313000027. 29. Sabatini R, Moore T, Hill C, A New Avionics Based GNSS Integrity Augmentation System: Part 2 – Integrity Flags, Journal of Navigation (Royal Institute of Navigation), vol. 66, vo. 4, pp. 511-522, June 2013. DOI: 10.1017/S0373463313000143. 30. Sabatini R, Moore T, Hill C, Avionics-Based GNSS Integrity Augmentation for Unmanned Aerial Systems Sense-and-Avoid, 27th International Technical Meeting of the Satellite Division of the Institute of Navigation: ION GNSS+ 2014, Tampa (Florida), September 2014. 31. Ramasamy S, Sabatini R and Gardi A, Avionics Sensor Fusion for Small Size Unmanned Aircraft Sense-and-Avoid, IEEE Metrology for Aerospace 2014 (MetroAeroSpace 2014), pp. 271-276, Benevento (Italy), May 2014. DOI: 10.1109/MetroAeroSpace.2014.6865933 32. Sabatini R, Gardi A, Richardson MA, LIDAR Obstacle Warning and Avoidance System for Unmanned Aircraft, International Journal of Mechanical, Industrial Science and Engineering,

International Science Index, vol. 8, no. 4, pp. 6273, April 2014. 33. Sabatini R, Gardi A, Ramasamy S, Richardson MA, A Laser Obstacle Warning and Avoidance System for Manned and Unmanned Aircraft, IEEE Metrology for Aerospace 2014 (MetroAeroSpace 2014), pp. 616-621, Benevento (Italy), May 2014. DOI: 10.1109/ MetroAeroSpace.2014.6865998 34. Rodríguez L, Sabatini R, Gardi A, Ramasamy S, A Novel System for Non-Cooperative UAV Sense-And-Avoid, European Navigation Conference 2013 (ENC 2013), Vienna (Austria), April 2013. 35. Ramasamy S, Sabatini R, Gardi A, Kistan T, Next Generation Flight Management System for Real-Time Trajectory Based Operations, Applied Mechanics and Materials, vol. 629, Trans Tech Publications, Switzerland, pp. 344-349, 2014. DOI: 10.4028/www.scientific.net/AMM.629.344 36. Ramasamy S, Sangam M, Sabatini R, Gardi A, Flight Management System for Unmanned Reusable Space Vehicle Atmospheric and Reentry Trajectory Optimisation, Applied Mechanics and Materials, vol. 629, Trans Tech Publications, Switzerland, pp. 304-309, 2014. DOI: 10.4028/www.scientific.net/AMM.629.304 37. Gardi A, Sabatini R, Ramasamy S, Kistan T, Real-Time Trajectory Optimisation Models for Next Generation Air Traffic Management Systems, Applied Mechanics and Materials, vol. 629, Trans Tech Publications. Switzerland, pp. 327-332, 2014. DOI: 10.4028/www.scientific.net/AMM.629.327 38. Gardi A, Sabatini R, Ramasamy S, de Ridder K, 4-Dimensional Trajectory Negotiation and Validation System for the Next Generation Air Traffic Management, AIAA Guidance, Navigation & Control Conference 2013 (GNC 2013), Boston, Massachusetts (USA), August 2013. DOI: 10.2514/6.2013-4893 39. Ramasamy S, Sabatini R, Gardi A, Liu Y, Novel Flight Management System for Real-Time 4Dimensional Trajectory Based Operations, AIAA Guidance, Navigation & Control Conference 2013 (GNC 2013), Boston, Massachusetts (USA), August 2013. DOI: 10.2514/6.2013-4763