5th EUROPEAN CONFERENCE FOR AEROSPACE SCIENCES

Robotic Relative Motion Reproduction for Air to AirRefuelling Simulation

P. R. Thomas?, T. S. Richardson? and J. L. du Bois???Department of Aerospace Engineering, University of Bristol, University Walk, Bristol, BS8 1TR

??Department of Mechanical Engineering, University of Bath, Bath, BA2 7AY, UK

AbstractHybrid testing involves the testing of physical subassemblies in a real-time simulation, coupled to numeri-cal simulations of complete systems and their related environmental conditions. It offers the ability to testthe performance of components in a realistic operating environment while remaining within the low-cost,repeatable, and relatively safe confines of a laboratory. The University of Bristol and Cobham MissionEquipment have developed such a hybrid test facility for relative motion simulation between two indepen-dent bodies. This paper describes the use of the facility in replicating an air-to-air refuelling environmentand how key latency issues were addressed.

1. Introduction

Air to air refuelling (AAR) is an important capability to extend the range, payload, and endurance of aircraft. Whilethere has been no interest in adopting AAR for civilian aviation there are a few proponents of its application,1 promotingreductions in fuel consumption for passenger and freight transport, the reduction of airport loading, the extension ofrange and payload in existing aircraft, and increased scope for scientific and environmental surveys through improvedendurance. For unmanned aerial systems (UAS), where endurance is no longer limited by pilot fatigue, aerial refuellingcapabilities offer significant benefits. Refuelling operations have historically been conducted by human pilots whorequire a high level of training and fast response times, and as such can not (in the near future) be conducted forremotely piloted aircraft over existing data links. The recent proliferation of UAS (including combat systems) hastherefore resulted in a demand for automated air-to-air refuelling (AAAR) capabilities and it is this requirement thatmotivates the work described in this paper. Successful accomplishment of AAAR relies on the development of two keytechnologies: Firstly, position sensing and tracking, to determine the relative position between aircraft and refuellingequipment; and secondly, control strategies to enable robust and safe operation of the aircraft in steering them to theirtarget.

The development of these two technologies relies on sophisticated testing: the sensor development requiresphysical tests under realistic conditions, while the control algorithm development leans heavily on realistic sensor datato ensure robust operation. It is advantageous to perform as many of these tests, and recreate as realistic conditions aspossible, in a laboratory environment. This permits the testing of sensors and systems critical to the safety of equipmentand personnel with reduced risk, and facilitates stage-gate management of large projects to mitigate financial risks. Thework described here is concerned with creating both a laboratory model-in-the-loop (MiL) test facility that can satisfythese requirements, and a simulated refuelling environment that is a sufficiently representative for such testing, inorder to provide the most comprehensive evaluation possible for these new aerial refuelling technologies prior to flighttesting.

Often this type of work will be related to sensing requirements, with no direct contact between the two bodies.A prominent example is the case of satellite docking approach, and it is unsurprising that some of the first largescale relative motion hybrid testing experiments have been focused on this problem.2–5 The system dynamics in anAAR context have much shorter timescales than satellite manoeuvres and the relative motion is more erratic due tothe stochastic nature of atmospheric turbulence and aerodynamic coupling. Robots have been used in other works tosimulate aircraft motion in refuelling operations; for example Pollini et. al.6 used a robot to recreate aircraft motion totest vision system algorithms, but the aircraft control loop was not closed.

The facility discussed in this paper uses two industrial robotic arms to manipulate refuelling hardware, drivenby a numerical simulation of the AAR scenario. The assembly of such a simulation model is no trivial task.7, 8 The‘hook-up space’ is a relatively compact environment with complex interactions between aircraft, refuelling equipment,

Copyright© 2013 by the University of Bristol. Published by the EUCASS association with permission.

FLIGHT DYNAMICS/GNC AND AVIONICS

and aerodynamic effects from the receiving aircraft’s bow wave, the wake vortices emanating from the tanker, and airturbulence. The model described in this paper provides a simulated environment for testing and evaluating a numberof different control approaches and designs, including autonomous control logic. Then, by incorporating prototype orproduction sensors into the facility, the measurable quantities can be fed back into the simulation model to close thecontrol loop and provide realistic tests in a controlled environment

The first part of this paper describes the MiL testing facility and discusses the control interfaces to achieve thenecessary response performance. The second part discusses the AAAR simulation environment which has been devel-oped to replicate the refuelling environment. Results from running the simulation on the facility are then presented.

2. Relative Motion Facility

The relative motion robotics (RMR) facility is comprised of two 6 degree-of-freedom (6DOF) robotic manipulators,one of which is mounted on a linear track. Full scale refuelling hardware is mounted on the robots, and a large range ofrelative motion can be accommodated to simulate the final 10 m of an approach in an aerial refuelling procedure. Therobotic manipulators with the refuelling hardware are shown in Figure 1. The robotic cell comprises two ABB IRB6640production-line robots, designated R1 and R2, and are supplied with a proprietary IRC5 controller. R1 is secured tothe ground whilst R2 is mounted on the 7.7 m IRBT6004 track to permit translation of the robot base at a rate of1.6 ms−1. The robots are electro-mechanically driven and each has 6 joints capable of producing 6DOF motion in threeCartesian coordinates and three orientational axes, with accelerations up to 2g and velocities of more than 2 ms−1. TheIRC5 controller provides a high level of functionality, including coordinate transformations, kinematic computations,and motor control loops which take account of the robot geometry, inertia and payload information. Importantly, theproprietary controller also provides several layers of safety controls to protect operators and equipment. A systemdiagram of the RMR facility is shown in Figure 2.

In normal (factory default) operation the user would program the robots using a high-level language calledRAPID. The instructions used in the RAPID code provide a powerful tool for quickly generating complex motionpaths and creating loops and conditional operation patterns. This language facilitates a variety of input and output(I/O) methods, including analogue and digital I/O channels as well as Ethernet communication protocols. The datasent and received on these channels can be used in the RAPID code in order to control the operation of the robots.In our bespoke MiL operation, simulation models are executed in real-time on a National Instruments PXIe-8133RT1.73 GHz control board mounted in a PXIe-1033 chassis. The PXIe system also runs a supervisory process responsiblefor orchestrating the test procedure and pre-processing of position demands sent to the robots. This supervisory processalso acts as a hub for all the signals associated with the test. The position demands are sent via communications threadson the PXIe to the robot control hardware. During either normal of MiL operation, the RAPID interpreter executesmotion instructions which pass the motion commands to a motion planning routine. This planning routine performskinematic computations and sends the joint motor demands to the axis computer.

Figure 1. Plan view of robot cell, depicting aerial refuelling hardware and coordinate system.

!"#"$%&'()#'*$+&(,

-.('$/

!&01!2

3'*(%&(,)$+.%"

!4

5'66+"

78'0"

3'*(%&(,)$+.%"

98.#:

Figure 2. Photograph of robot cell, showing aerial refuelling hardware mounted on robots.

able commercial equipment, looking at three di!erent control approaches and describing their merits anddrawbacks. Section III presents a high level interface approach and examines the achievable performance.Addressing shortcomings in this approach, Section IV describes preliminary tests of a low level interface andsets out important safety considerations for a facility of this scale. Section V discusses optimisation of themotion paths, looking at peculiarities of the physics of robot arm manipulators in this type of application.Section VI then presents the flight dynamics simulation for use in the aerial refuelling tasks and showsthe preliminary performance results for the facility. Conclusions are drawn and future work is discussed inSection VII.

II. Control Topologies

The robots used in the Bristol University relative motion robotic facility are ABB IRB6640 production-linerobots and are supplied with a proprietary IRC5 controller. The robots are electromechanically driven andeach has 6 joints capable of producing 6DOF motion in three Cartesian coordinates and three orientationalaxes, with accelerations up to 2g and velocities of more than 2m/s. The IRC5 controller provides a highlevel of functionality, including coordinate transformations, kinematic computations and motor control loopswhich take account of the robot geometry, inertia and payload information. Importantly, the proprietarycontroller also provides several layers of safety controls to protect operators and equipment.

In normal operation, the user would program the robots using a high-level language called RAPID code.

3 of 21

American Institute of Aeronautics and Astronautics

Figure 1: The relative motion robotics facility mountedwith refuelling hardware

MODEL EXECUTION

RELATIVE MOTION ROBOTIC FACILITY

SE

NS

OR

S

RO

BO

T C

ON

TR

OL

LE

R

SIMULATION

MODEL

VISUALISATION

REAL-TIME PXI

PLATFORM

ORCA

COMPUTER

Figure 2: Layout of the facility as a synthetic environmentfor replicating air to air refuelling

2

P. R. Thomas et. al. ROBOTIC RELATIVE MOTION REPRODUCTION FOR AAR SIMULATION

3. Control Interfaces

Three control interfaces for the input of motion demands from the simulation to the robot controller were initiallyconsidered:

1. high level (through RAPID)

2. mid level (exploiting force control inputs and using auxiliary feedback)

3. low level (direct access to axis computer of robot controller)

The biggest advantage of the first two options is that they retain the robust safety mechanisms of the proprietarycontroller, and thus permit faster development of auxiliary control without the fear of serious malfunction and injury ordamage. The high level option also retains the matured control technology of the proprietary controller, providing thebest motion path control for the least development effort. However this is provided at the cost of deterministic real-timecontrol; the RAPID code is not intended to receive, parse, and compute small motion path segments on the fly in thismanner. The mid level control option uses the input signals normally used for force control feedback to affect themotion of the robots. In this manner the RAPID interpreter and much of the motion planning algorithm is bypassed,resulting in a much faster control loop whilst the safety of the proprietary controller is preserved, as are the kinematiccomputations and coordinate transformations. The specific drawback of this method is that the new feedback controlmust be tuned and will not easily achieve the same performance as the inner loop proprietary control. The third andfinal option is to directly access the axis computer of the robot controller. Whilst offering the best real-time controlof the robots, adopting this approach requires significant development effort, and there can be a reduction from theintrinsic accuracy of the industrial controller. A further consideration is that this technique undermines some of themore sophisticated elements of the safety controller, and alternative safeguards need to be implemented.

The control paradigm chosen for the RMR combines two of the interfaces discussed above. In this way thebenefits of each are largely preserved and the drawbacks predominantly avoided. The high level interface (RAPID)is used to define a nominal trajectory, and the low level interface (ORCA) is used to compensate for delays whileremaining within appropriate margins of the known safe trajectory.

3.1 RAPID interface

The biggest barrier to implementing a real time scheme using this approach lies in the non-determinism of the com-munication protocols and the unpredictable nature of the RAPID code interpretation. The former is imposed by theimplementation of TCP/IP Ethernet communications on the robot controller. The unpredictable nature of the RAPIDcode is thought to be due to the fact that it is being interpreted on a processor running a variety of concurrent threadsso execution can slow down when the processor is heavily loaded. In normal operation this is not perceptible but whenpositions are being demanded at rates of 10 Hz or more the system is sensitive to small delays in the execution cycle.Safety mechanisms on the controller are also thought to play a part and the controller will halt the motion if it cannotmeet certain stopping distance criteria. Methods for mitigating these effects and providing a real time deterministicmotion based on the deterministic outputs of the numerical simulation are required.

The flow of information between the simulations and the robots is illustrated in Figure 3. The PXIe, the IRC5controller, and the robots are shown as three physical devices and the distinct processes and threads running on thedifferent devices can be seen. The important elements to note are the buffers on the PXIe and the IRC5 controllerwhich form the gateways between the deterministic execution at either end of the diagram and the non-deterministicmessage processing in the centre of the diagram. These are necessary to ensure determinism but introduce undesirabledelays.

3.2 ORCA interface

In order to facilitate low-level control of the robot hardware, the Open Robot Control Architecture (ORCA) of theUniversity of Lund9 has been adopted. This control uses a separate ORCA PC which intercepts signals sent betweenthe main computer and the axis computer in the IRC5 controller. It can then augment or override the signals sent tothe axis computer and demand joint motor positions and velocities directly. ABB use position demands in conjunctionwith velocity and torque feed forward to produce accurate control of the robots. The torque signals are consideredcommercially sensitive, and are disabled by ABB as part of the licensing agreement for the ORCA interface, leavingthe position and velocity feed forward available for use through ORCA. In addition, the controller gains can also betuned through ORCA, raising the possibility of gain scheduling.

3

FLIGHT DYNAMICS/GNC AND AVIONICS

Figure 3. Position and control data flow between processes on the PXIe (real time) and IRC5 (proprietary robot) controllers

6of21

Am

erican

Institu

teofA

eronautics

and

Astro

nautics

Figure 3: Position and control data flow between processes on the PXIe and IRC5 controllers.

!"

"!"#$%&'()('*+,-$.'-/'.0&'1-23"4&5'6789:);6<7'"4.&%/+1&'1-4/"#$%+."-4'

""

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

F4G!" !"#$ !% !%#$ &' &'#$

'

$'

!''

!$'

&''

&$'

()*+,-./

()*+,0)(1)2,3435+,-*./

,

,3435+6)*+*78+92.(:;3()72*+..<=+><:.+*+..<=+?+3+)8+*+..<=+@+2A*+..<=+B7*C7.+

" F,G "!"#$%&'()>'?&0+@"-$%'-/'.0&'2-."-4'AB+44"4#'+C'.0&'2-."-4'1-22+45C'+%&'1-22&41&5'

.0%-$#0'.0&'6789:'"4.&%/+1&D'E+F'G.+1H&5'3+%'10+%.'C0-I"4#'1-22$4"1+."-4'1,1B&'."2"4#J'E3F'<-$4.&%C'C0-I"4#'.0&'4$23&%'-/'2-."-4'"4C.%$1."-4C'A%-1&CC&5'"4'.0&'6789:'1-5&'+45'

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

Figure 4: Layout of the combined RAPID-ORCA interface.

The biggest problem introduced with the use of the ORCA interface is that by directly passing demands to theaxis computer, much of the robust safety intrinsic to the industrial control systems is bypassed. To minimise thisrisk, the approach adopted here is to use the ORCA interface only to augment the control of the robots. The high-level RAPID interface remains as the primary input to the robot control, with the ORCA interface used to augmentthe position and velocity in order to compensate the delay in the high-level control. The extent to which the ORCAinterface can modify the signal from the IRC5 main computer is therefore limited, ensuring the robots do not deviatesignificantly from the safety-assured path determined by the main IRC5 controller. The layout of this system is shownin Figure 4. The approach presented can therefore produce fast system response times without forsaking the robustsafety of the high-level approach.

3.3 Performance

Figure 5 shows the step response of the robots to a Cartesian position demand. This demand is in the x direction forR2, and uses all six robot joints as well as the track motion. Three configurations are used: First, the response usingonly the RAPID interface, with a motion command rate of 5 Hz. Second, the RAPID interface in conjunction withaugmented position demands through the ORCA interface. Third, the RAPID interface in conjunction with augmentedposition and velocity demands through the ORCA interface.

The latency in the RAPID interface is clear, with more than half a second passing before any motion commences.The motion then exhibits a small overshoot and slowly approaches the reference signal. In contrast, both of the ORCAinterface examples show a very fast response to the demand. The augmented position curve appears to be producinga similar overshoot to the first example, but because the velocity demand from the RAPID interface is unmodified,this produces a late acceleration, explaining the large overshoot that follows. This curve then converges on that of theRAPID interface. The final example, using both position and velocity augmentation in the ORCA interface, shows adramatic improvement in the initial response, where the velocity is close to matching the step of the reference signal.There is then a rebound, presumed to be due to the inertia and elasticity of the robots, and the slower position controlloop brings the signal back in line with the reference, again with some overshoot. In practice, a step response in theposition demand is an unrealistic criterion for the motion of a physical system, so the rebound seen in the final exampleis not a concern. It is encouraging that the initial response shows minimal latency, measured to be in the region of

4

P. R. Thomas et. al. ROBOTIC RELATIVE MOTION REPRODUCTION FOR AAR SIMULATION

!"

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

1 2 3 4 5 6

0

0.2

0.4

0.6

0.8

1

1.2

1.4

time (s)

disp

alce

men

t (m

m)

referenceRAPID+ORCA(pos,vel)RAPID+ORCA(pos)RAPID

"!"#$%&'()*'+,&-'%&.-/0.&'1/%'2'32%,&."20'4)5"%&6,"/0'5&7205'

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

Figure 5: Robot step-response performance.

20 ms for the initial response and a short rise time, making the complete system suitable for high fidelity non-contactMiL simulations.

4. Air to Air Refuelling Simulation

Simulations are written in Simulink and compiled for use on the PXIe platform using National Instruments’ Veristandtarget language compiler. The simulation model (see Figure 6) takes into account:

• Aircraft (tanker and receiver) trajectory demands, navigation logic, and flight control systems.

• Dynamics models for the aircraft, hose, and paradrogue assembly.

• Atmospheric (gust and wake) and bow wave disturbance models

Both the receiver and tanker are rigid-body, six degrees of freedom objects having nonlinear aerodynamic be-haviour in the form of lookup data. Reference commands from the guidance and navigation systems are used by eachaircraft’s flight control system to generate input commands to the aircraft. These in turn, along with the dynamic air-craft states, are used to generate the aerodynamic forces and moments on the aircraft. In addition to the forces andmoments from engine thrust and gravity, the total forces and moments are then used to recurrently solve the equationsof motion. The guidance and navigation systems for the tanker define a fixed-heading or racetrack manoeuvre for thetanker to follow, whilst the receiver uses a waypoint-based navigation system controlled by a finite state-machine. Bothaircraft have lower level control and stability augmentation systems, appropriate to their type.

Additional intermittent forces and moments on the aircraft come from atmospheric turbulence which is charac-terised by random, homogenous, and isotropic behaviour. It can be modelled well by passing white noise with unityspectral density through a low-pass shaping filter that gives the desired output spectrum. The continuous Dryden-formfor the filters are used since they possess rational power spectral densities making their modelling far simpler.10 Theaerodynamic model for the wake vortices is based on the work by Saban, Whidborne, and Cooke.11 The induced ve-locity generated by a finite wing is modelled using Weissinger’s lifting line theory in combination with Kurylowich’svortex model. The Weissinger lifting line model is the simplest vortex panel method. It works well for swept wingsand converges to reasonably good solution in both low and high aspect ratio limits. The Kurylowich model on the otherhand is simple and effective in merging the individual vortices generated by a finite wing into a pair of counter-rotatingwingtip vortices. A section through the wake flow field at 30 m distance behind the tanker is shown in Figure 7.

The trailing hose and drogue is modelled as a series of lumped-mass, rigid cylindrical links connected by fric-tionless ball-and-socket joints, after the work by Ro et. al.12, 13 Link masses, and gravitational and aerodynamic forces,are lumped at the joints, and drogue mass and forces, modelled with a parametric drag equation, are applied to the endof the final link. The turbulence effects applied to the aircraft are similarly applied to the hose and drogue. The bowwave preceding the receiver, which additionally upsets the drogue’s flight, is approximated with the flow generatedaround a three-dimensional Rankine half-body. The behaviour of a hose drum unit is also modelled to reel in and outthe hose based on the tension modelled in the hose linkages.

5

FLIGHT DYNAMICS/GNC AND AVIONICS

Sequence

logic

Tracking

algorithm

State demand

generator

Control

demand

generator

Flight

dynamics

model

Probe

model

Hose

dynamics

model

Air

turbulence

Wake

turbulence

Bow wave

model

Trajectory

generator

State demand

generator

Control

demand

generator

Flight

dynamics

model

Drogue

dynamics

model

TURBULENCE

TANKER

RECEIVER

Drogue states

Tanker states

Probe states

Receiver states

Figure 6: System diagram of the air to air refuelling simulation model

surfaces. Ideally a strip element theory method could be usedto discretise the receiver wing surface and calculate the indi-vidual lift and drag components on each strip individually how-ever such an approach is computationally expensive and wouldmake real-time simulation difficult. The induced velocity fieldis approximated around the CM of the receiver by averaging thewind components and gradients over the receiver lifting surface[36, 37]. The velocity induced on the receiver by the tanker iscomputed at each calculation point (along the 1/4 chord line) ofthe receiver using (17). The induced velocity field thus obtainedis highly non-uniform, and can be approximated around the CMof the receiver vehicle as the sum of uniform wind componentsand uniform wind gradients.

As an example, the wake vortex field is calculated for an an-gle of attack, α = 2.8◦, zero sideslip and an indicated airspeedV of 200 m/s. The resulting streamline plot is shown in Fig-ure 6(a). Sections through the wake flow field at increasingdistance from the tanker are shown in Figure 7. The jetwashfield is shown in Figure 6(b).

Is there any verification against CFD data?

2.6. Boom model

An empirical model of a flying boom was developed as partof a series of US Air Force Research Laboratory (ARFL) re-fuelling studies [38]. These models featured a non-retractablerigid body with rotation about the boom root which translatedthrough the air, with variable mass and inertia to represent theeffect of the retractable boom, physical geometry, aerofoil data,and component weights [39]. Aerodynamic, gravitational, andcontrol input terms were applied directly to equations whichcomputed the moments about the boom root joint. More re-cently Smith and Kunz [7, 40] derived a new set of multi-bodysystem equations and aerodynamic terms to characterise thefixed and extendable parts of the boom, before coupling themwith a tanker model. The data from these model has been usedextensively in other works [8, 41]. On the other hand Vendraet. al. [42] favoured a 3D finite element model (FEM) to gener-ate the boom behaviour in a form typically used to model robotmanipulator kinematics. Doing so enabled certain elastic be-haviour to be captured in the model.

The boom model described herein is taken from [40] but us-ing a different description for the boom. However unlike theirmodel, but similar to the other works above, we only model theboom with regards to the tanker via a fixed tow point for ref-erence. This is acceptable since we do not currently considerlarge dynamic motion of the tanker in our refuelling simula-tion. Consider the illustration in Figure 8 showing the flyingboom and telescopic extension. The declination and azimuth ofthe boom are denoted θ f and ψ f and are shown in their positivesense. These follow the standard notation given the orientationof the flying boom axes frame. The length of the flying booml f is fixed at 11.28 m, whilst the telescopic probe extension canrange between 0 ≤ le ≤ 6.4 metres.

The equations of motion for the boom can be described interms of the dynamics of both the flying boom ( f ) and the tele-scopic extension (e), with respect to their own axes frames, in

5.2 Example Flows

As an example, the wake vortex field is calculated for an angle of attack, α = 2.8◦, zero sideslip and anindicated airspeed V = 200 m/s. The resulting streamline plot is shown in Figure 24. Sections throughthe wake flow field are shown in Figure 25. Note that the the axes system is in the B frame, so positivez is down. The additional velocity induced by the added fuselage elements is shown in Figure 26 at aposition x = −30 m.

The jetwash field is shown in Figure 27.

Figure 24: A330-200 Streamlines

28

(a) Vortex streamlines

Figure 27: A330-200 Jetwash Flow Field

6 A400M Models and Results

6.1 A400M Main Wing Discretization

The geometric characteristics of the A400M is given in Table 3. The fuselage is included as a secondlifting surface as shown in Figures 28 and 29. The plan view is shown in Figure 30. The fuselage/mainwing surface with a wire model of the A400M overlaid upon it is shown in Figure 31.

Table 3: Geometric parameters of A400M main wing.

Parameter Main Wing FuselageAspect ratio, AR [–] 8.1 0.2Wing span, b [m] 42.4 4.0Wing sweep, Λ1/4 [◦] 15.0 0Taper ratio, λ [–] 0.345 0.9Dihedral, Γ [◦] -2.0 0Angle of incidence, α0 [◦] 0 0Wing twist, αtwist [◦] 0 0number of segments, Nseg 12 0

30

(b) Jet wash

Figure 6: Tanker flow fields.

40 30 20 10 0 10 20 30 4030

20

10

0

10

20

30

y

z

x = 40 m

40 30 20 10 0 10 20 30 4030

20

10

0

10

20

30

y

z

x = 30 m

40 30 20 10 0 10 20 30 4030

20

10

0

10

20

30

y

z

x = 20 m

40 30 20 10 0 10 20 30 4030

20

10

0

10

20

30

y

z

x = 10 m

h (m)

y (m)y (m)

y (m) y (m)

h (m)

h (m)

h (m)

x = 10 (m)x = -20 (m)

x = -30 (m)x = -40 (m)

Figure 7: A330-200 wake vector flow field sections (shown with respect toheight, h = −z, for clarity).

7

Figure 7: Typical wake vortex field section

The simulation structure is purposely modular such that ongoing improvements to individual componentscan be made in parallel and swapped in, limiting the changes needed to the simulation environment.

The purpose of the simulation, in the context of the RMR facility, is to generate position and orientationinformation for the probe and drogue which can be replicated by the manipulators. To that end we definea set of axes systems in Figure 12 which identifies the refuelling probe (p) and paradrogue (d) objects. Thetask in probe-drogue configured AAR is to approach and couple the probe with the drogue to close the refuelline. Consequently the probe must track and close the range between it and the drogue, this is describedin terms of the approach frame (a) which is coincident with the drogue. The probe position is thereforedescribed with the coordinates (xp, yp, zp), relative to the origin oa.

!!

"!

#!

#"

""

!"

$#

%&'&

(&$"&)*#&)*!&%# "#

##

!#

Figure 12. Probe (p), drogue (d), and approach (a) axes definitions.

VI.A. Aircraft models

Both the receiver and tanker are rigid-body, six degrees of freedom objects having nonlinear aerodynamicbehaviour in the form of lookup data. The general schematic for the rigid bodies is illustrated in Figure13. Reference commands from the guidance and navigation systems are used by the flight control system togenerate input commands to the actuator models. These in turn, along with the dynamic aircraft states areused to generate the aerodynamic forces and moments on the aircraft at the centre of mass (CM). Clearly theCM will vary throughout the refuelling process, primarily a!ecting the pitching moment of both receiver andtanker. However up to now we have assumed the variation will have a negligible e!ect on the performance ofthe flight control laws and have used a fixed CM at 0.25c i.e. 25% from the leading edge of the wing’s meanaerodynamic chord. Future improvements to the simulation will determine if this was a valid assumption: ithas already been suggested that that mass variation due to fuel transfer compounds the di"culties createdby tanker wake turbulence.45 A generic tanker flight dynamics model is employed but the tanker dynamicsare not critical to the simulation - in simpler scenarios the tanker model has been replaced with a referencepoint moving at constant velocity. Two configurations for the receiver aircraft are used: an F-16 fighter jetand the conceptual Innovative Control E!ector aircraft.

A model for an F-16 unmanned jet fighter was derived from the data in,46 which itself is a reduced versionfrom.47 The simplified model is valid for the aerodynamic range ! ! ["10!, 45!], " ! ["30!, 30!], which iswell within the flight regime for refuelling aircraft. Three first order lags with rate limits and saturationsmodel the actuators similar to those used in.47 Aerodynamic forces and moment coe"cients about the centreof mass (CM) are calculated in the aerodynamic subsystem using the previous time step aircraft states:

CX(!, q, #e) CY (!,", p, r, #a, #r)CZ(!,", q, #e)

CL(!,", p, r, #a, #r) CM (!, q, #e, CZ)CN (!,", p, r, #a, #r)

where !," are the aerodynamic incidence and sideslip angles and p, q, r are the rotational rates. Theparameters #e, #a, and #r correspond to the elevator, aileron, and rudder deflections. Leading edge flaps anddi!erential tail inputs are not used in the model. The propulsive thrust is calculated through a lag in thepower generated by the jet engine simulated with a first order transfer function.

A model for the conceptual Innovative Control E!ector (ICE)48 aircraft is used in addition to the F-16to investigate control challenges relevant to future aircraft configurations. The ICE is a tailless delta wing

13 of 21

American Institute of Aeronautics and Astronautics

Figure 8: Probe (p), drogue (d), and approach (a) axesdefinitions

The output of the simulation for its use in the RMR is the position and orientation information for the probeand drogue which is replicated by the manipulators. To that end we define a set of axes systems as in Figure 8 whichidentifies the refuelling probe (p) and paradrogue (d) objects. The task in probe-drogue AAR is to approach and couplethe probe with the drogue. Consequently the probe must track and close the range between it and the drogue, this isdescribed in terms of the approach frame (a) which is coincident with the datum – the location of the drogue whentrailed from the tanker, in still air. The probe and drogue positions are then described with the coordinates (xp, yp, zp)and (xd, yd, zd), relative to the origin oa.

6

P. R. Thomas et. al. ROBOTIC RELATIVE MOTION REPRODUCTION FOR AAR SIMULATION

5. Real-time MiL Simulation

For the MiL tests the probe and drogue motion are made relative to the drogue’s datum point, and are reproducedindependently by R2 and R1 respectively. One degree of redundancy remains: the track motion. This is resolvedby separating the motion of the probe into high and low-frequency components; the robot axes are used to performthe high-frequency motion and the track moves the robot base to provide the low-frequency, quasi-static responseand give the probe its full longitudinal operational range. Figure 9a shows the absolute position data, including the3DOF translational position output from the simulation overlaid with the measured response of the robots. At thisscale the lines appear coincident. The simulation shows a position hold approximately 5 m aft of the drogue, followedby an approach to the pre-contact position, which is again held approximately 2 m aft of the drogue, and finishingwith an aggressive engagement. The simulated refuelling procedure is conducted in light turbulence. Figure 9b showsthe position error from these plots. In making this comparison the time base of the measured signal was adjusted toaccount for the delay, with the intention of presenting position errors rather than temporal errors. It is interesting tonote that any effects from corner path artefacts are indiscernible in the presence of other disturbances. The large peakat approximately 30 s corresponds with the rapid approach of the receiver aircraft to the pre-contact position. A similarsharp rise is seen at the end of the plot where the final engagement is made. Although these errors are presented hereas positional errors, they are found to be better described as temporal discrepancies; the differences seen are the resultof a lag between the demanded motion and the measured robot position when moving at high speeds. What is notapparent in this figure, but can be determined from close examination of Figure 9a, is that while the robot motion lagsthe demand at some points, it leads the demand at others. This may point to a variable frequency response that couldbe characterised to the ends of further improving the performance using a feedforward control approach.

6. Conclusions

An overview of the Relative Motion Robotic (RMR) facility at the University of Bristol, developed in collaborationwith Cobham Mission Equipment, has been given and the important considerations in implementation and performancehave been discussed. Timing is critical in structural hardware in the loop simulations, and factors affecting performancein this regard have been described. Steps taken to improve the performance and to push the limits of the equipmentbeing used were discussed. The outcome of this work has been a facility to implement high-bandwidth, closed-loop,hybrid simulations in a robust, stable, and realistic manner. A high fidelity multi-entity simulation model has beendeveloped and sample results from a simulated air to air refuelling exercise on the facility have been presented. Theseresults demonstrate the suitability of the RMR for conducting advanced tests of aerial refuelling hardware and sensors,for the purpose of developing automated aerial refuelling capabilities. This will lead to tests of novel sensing andcontrol technologies for autonomous air-to-air refuelling. Beyond this the focus will be shifted towards including forcefeedback capabilities and developing methods for simulating discontinuous contact events.

0 10 20 30 40 50 60−1000

0

1000

2000

3000

4000

5000

6000

time (s)

posi

tion

(mm

)

x2simy2simz2simx2measy2measz2meas

(a) Absolute positions from the simulated data and fromthe measured robot positions.

0 10 20 30 40 50 60−20

0

20

40

60

80

100

time (s)

posi

tion

erro

r (m

m)

x2y2z2

(b) Position error for the robot motion relative to the pre-scribed simulation data motion path.

Figure 15. Positional data from the real time flight simulation.

Figure 16. Results of the real time simulation running on the robots

18 of 21

American Institute of Aeronautics and Astronautics

(a) Absolute positions

0 10 20 30 40 50 60−1000

0

1000

2000

3000

4000

5000

6000

time (s)

posi

tion

(mm

)

x2simy2simz2simx2measy2measz2meas

(a) Absolute positions from the simulated data and fromthe measured robot positions.

0 10 20 30 40 50 60−20

0

20

40

60

80

100

time (s)

posi

tion

erro

r (m

m)

x2y2z2

(b) Position error for the robot motion relative to the pre-scribed simulation data motion path.

Figure 15. Positional data from the real time flight simulation.

Figure 16. Results of the real time simulation running on the robots

18 of 21

American Institute of Aeronautics and Astronautics

(b) Relative position error

Figure 9: Positional data from the real time flight simulation

7

FLIGHT DYNAMICS/GNC AND AVIONICS

7. Acknowledgments

The authors wish to express their gratitude to Anders Robertsson, Anders Blomdell, and Klas Nilsson at the Universityof Lund, Sweden for the considerable time they have spent helping with the setup of the low-level ORCA interfacewith the ABB IRC5 controller. The work described here has been largely funded by Cobham Mission Equipment aspart of the ASTRAEA programme, which seeks to enable the routine use of UAS (Unmanned Aircraft Systems) in allclasses of airspace without the need for restrictive or specialised conditions of operation. The ASTRAEA programme isco-funded by AOS, BAE Systems, Cobham, EADS Cassidian, QinetiQ, Rolls-Royce, Thales, the Technology StrategyBoard, the Welsh Assembly Government and Scottish Enterprise.

References

[1] Nangia, R. K., Operations and aircraft design towards greener civil aviation using air-to-air refuelling, The Aero-nautical Journal, 110(1113):705–721, 2006.

[2] Newhook, P. and Eng, P., A robotic simulator for satellite operations, in Proceedings of the 6th InternationalSymposium on Artificial Intelligence, Robotics and Automation in Space (i-SAIRAS), Montreal, Canada, June2001.

[3] Kaiser, C., Rank, P., Krenn, R. and Landzettel, K., Simulation of the docking phase for the SMART-OLEVsatellite servicing mission, in Proceedings of the 9th International Symposium on Artificial Intelligence, Roboticsand Automation in Space (i- SAIRAS), Hollywood, California, USA, 2008.

[4] Boge, T., Wimmer, T., Ma, O. and Zebenay, M., EPOS–A robotics-based hardware-in-the-loop simulator forsimulating satellite RvD operations, in Proceedings of the 10th International Symposium on Artificial Intelligence,Robotics and Automation in Space (iSAIRAS), Sapporo, Japan, August–September 2010.

[5] Suatoni, M., Mollinedo, L., Barrena, V., Colmenarejo, P. and Voirin, T., Use of COTS robotics for on-ground val-idation of space GNC systems: Platform dynamic test bench, in Proceedings of the 11th International Symposiumon Artificial Intelligence, Robotics and Automation in Space (i-SAIRAS), Turin, Italy, September 2012.

[6] Pollini, L., Mati, R. and Innocenti, M., Experimental evaluation of vision algorithms for formation flight andaerial refueling, in AIAA Modeling and Simulation Technologies Conference and Exhibit, Providence, RhodeIsland, August 2004.

[7] Dogan, A., Sato, S. and Blake, W., Flight control and simulation for aerial refueling, in AIAA Guidance, Naviga-tion, and Control Conference and Exhibit, San Francisco, California, USA, August 2005.

[8] Wang, J., Patel, V. V., Cao, C., Hovakimyan, N. and Lavretsky, E., Novel L1 adaptive control methodology foraerial refueling with guaranteed transient performance, Journal of Guidance, Control, and Dynamics, 31(1):182–193, 2008.

[9] Blomdell, A., Dressler, I., Nilsson, K., Robertsson, A., and Dressler, I., Flexible application development andhigh-performance motion control based on external sensing and reconfiguration of ABB industrial robot con-trollers, in Proceedings of the ICRA 2010 Workshop on Innovative Robot Control Architectures for Demanding(Research) Applications, Anchorage, Alaska, May 2010.

[10] U.S. Military Specification, Flying qualitites of piloted aircraft, MIL-F-8785C, November 1980.

[11] Saban, D., Whidborne, J. and Cooke, A., Simulation of wake vortex effects for UAVs in close formation flight,The Aeronautical Journal, 113(1149):727–738, 2009.

[12] Ro, K., Basaran, E. and Kamman, J. W., Aerodynamic characteristics of paradrogue assembly in an aerial refuel-ing system, Journal of Aircraft, 44(3):963–970, 2007.

[13] Ro, K. and Kamman, J. W., Modeling and simulation of hose-paradrogue aerial refueling systems, Journal ofGuidance, Control, and Dynamics, 33(1):53–62, 2010.

8