SEAS DTC Second Conference Proceedings v2

Autonomous Soaring Project Phase 2 R. Irvinea, F. Innesa, A. Brownb, S. Vosperb, G. Rooneyb, B. Devenishb, M. Hookc, E. Sparksc a MBDA UK Limited, PO Box 5, Golf Course Lane Filton, Bristol, BS34 7QW b Met Office, Fitzroy Road, Exeter, Devon, EX1 3PB c Roke Manor Research Limited, Romsey, Hampshire, SO51 0ZN

Abstract This SEAS DTC funded research aims to demonstrate autonomous soaring using a surrogate manned sailplane. This phase of the research follows from the Innovation Funded research previously undertaken within the SEAS DTC. The demonstrations undertaken during this phase will show the potential capability to predict the soaring possibilities within the local atmosphere in order to improve the range and endurance of autonomous flight vehicles. Keywords: autonomous, soaring, orographic lift, thermal lift, range, endurance, AAV Introduction Phase 2 of the Autonomous Soaring Project follows on from the SEAS DTC Innovation Funded research undertaken last year. Phase 2 brings together three organisations: MBDA, Met Office, Roke Manor Research. The objectives of Phase 2 are to build upon the paper study undertaken previously and demonstrate the capability to predict and exploit soaring opportunities within the local atmosphere. The project explores orographic (ridge) lift and thermal lift. The former is based on a meteorological forecast of vertical air speeds at various altitudes above points on a chosen grid while the latter draws on two sources; observation and automatic interpretation of clouds via a digital camera and tables of probability of thermals based on land type. Note that it is equally important to predict regions of falling air in order to attempt to avoid them as it is to predict regions of rising air to exploit the soaring opportunity. Autonomous soaring is opportunistic in nature and has a very strong temporal aspect. However it is foreseen that exploitation of soaring energy within the atmosphere can be of benefit to a number of autonomous air vehicles (AAV).

Example military uses could include comms relay, observation platform, loitering munition or long-range weapon. An example civilian use could be an AAV tasked with undertaking a pipeline survey. Trial Site Selection To reduce the cost of the demonstration trials and remove the development cost of safety critical algorithms the decision was made at the start of Phase 2 to use surrogate-manned sailplanes rather than a fully autonomous sailplane; the surrogate sailplane pilot having full authority to abort the trial at any time. This necessitated the requirement to engage the services of a gliding club to provide the surrogate sailplane. A number of gliding clubs within the UK were considered. The site deemed most suitable for purpose was the Bristol and Gloucestershire Gliding Club (BGGC) located by Nympsfield in Gloucestershire. The site of the BGGC airfield is on the edge of the Cotswolds by Birdlip escarpment and overlooking the Severn estuary. This location offers some very good orographic lift as winds channelled up the Severn estuary are forced up and over the escarpment. The locality also offers numerous examples of terrain usage from estuary and lake to forest, open fields,

2nd SEAS DTC Technical Conference - Edinburgh 2007

C17

urban etc… These different terrains uses offer may opportunities for thermal activity. A visit to the BGGC provided lots of anecdotal evidence to support this.

from the project a number of shake down trials were undertaken. These initial flights proved the GPS total energy sensor, data modems and moving map functionality.

The BGGC were keen on supporting the Phase 2 activities. During the initial site visit a range of different single and twin seat sailplanes were presented as possible surrogate gliders.

Orographic Met Model

Initial discussions with the BGGC instructors suggested that a number of orographic and thermal soaring opportunities should be available within a 10 km radius of the launch site. It was agreed to have a trials radius of 20 km from the launch site at Nympsfield. Communications Layout Wherever possible COTS hardware and software would be utilised to facilitate the functions required for the Phase 2 trials. One element of undertaking the trials that early on in the Phase 2 planning was deemed unknown was the physical method of transferring data to and from the moving sailplane and what data would require transfer during the trials. After a number of iterations an outline communications layout as shown in Figure 1 was developed. Sailplane

Ground Station

GPS and TE Sensor

Aerial

Predictions of the atmospheric vertical velocity are made using a numerical model [1] for airflow over complex terrain. The model solves a set of equations (simplified in the interests of computational efficiency) which determines the vertical motion induced by the flow of air over the real terrain. Digital terrain data are used at the lower boundary of the model and vertical profiles of wind and temperature from the coarser resolution Met Office Global forecast model are used to represent the undisturbed atmospheric conditions at any given time. The model then predicts the vertical velocity field above the terrain as a gridded data set with a 250 m horizontal resolution. An example of such a prediction is shown in Figure 2. Predictions such as these have been compared with past weather balloon and aircraft measurements of vertical velocity fields over mountains in a variety of locations [1,2]. In general the predictions have been shown to accurately represent both the strength and position of the up and downdraughts.

Data Modem

Cloudscaping Data

Data Modem

GPS and Total Energy Sensor

Camera

GPS and TE Sensor

Laptop EW Recorder

Way Point Generation

GPS Signal

Atmosphere Data

Met Model

GPS and Waypoint File

Moving Map Visualisation Bearing Requirement to Next Waypoint

Sailplane Pilot

Bearing Requirement (via Voice Comms Link)

Ground Operator

Figure 1: Proposed Comms Layout Figure 2: Example Prediction of Orographic Lift

This communications and data flow structure provided an element of risk to the Phase 2 demonstrations. To reduce the risk

Figure 2 shows the predicted vertical velocity field (colour contours, units m/s) at 500m above sea level over the


C17

Gloucestershire hills on 19 March 2007, at 18 Z. Also shown are the terrain height contours (solid lines) and horizontal wind vectors. Thermal Met Model A surface energy balance model has been run, driven by two years of observed atmospheric data, to predict the temperature variations of a number of different surface types (e.g. grassland, woodland, tarmac, urban). The average predicted differences between the temperatures of the different surfaces have then been presented as a function of season, time of day and cloud cover. For example the average temperature of a tarmac surface on a cloud-free day in summer is found to be 8 degrees warmer than that of a grass surface, but the differential decreases with increasing cloud cover or in other seasons. Large-eddy simulations are now being performed to evaluate the likelihood of having a thermal above an isolated patch which is warmer than its surroundings. Both the temperature contrast and the size of the patch will be varied. The aim is to combine the information obtained from this study from that from the surface energy balance model, to produce probabilistic predictions of the likelihood of finding a thermal above a given surface. These probabilistic predictions are presented as tables of probability based on types of surface (e.g. grass, wood, road, lake, town) and parameters such as time of day and cloud cover. The tables will then be applied to a map of the geographic area of interest in order to identify potential regions of lift or sink to be taken into consideration during thermal soaring trial flights. Cloudscaping A cloudscaping technique will be used to compliment the probabilistic predictions from the Thermal Met Model.

There are a number of techniques already developed to detect thermals. These include both low technology methods such as thermal poles and bubbles (most commonly used in competitive model sailplane flying) and high technology approaches such a Doppler lidars. These techniques are likely to be severely limited in the range at which thermals could be detected. We aim here to develop a method that will allow thermals to be detected over a considerable area. A thermal may trigger cloud formation as the rising air draws water vapour above the lifting condensation level (the altitude where air temperature equals the dew point). The detection of new and/or growing cloud could therefore provide a reliable indication of rising air. As the thermal decays the period of cloud growth will end and the cloud will slowly disintegrate. This disintegration will be driven by dispersion of the cloud and also the air descending back to lower levels. This suggests that if tracking the evolution of clouds were possible then it may be possible to identify both areas of rising and falling air. Determining the evolution of clouds is complicated by the wind and the density of clouds. An isolated cloud will change its appearance due to changes in view direction and relative position in addition to changes due to evolution. The approach chosen is to track small regions of cloud over a sufficiently long time frame that it is possible to categorise the movements into those that may be explained by the wind and those which cannot. As the cloud density increases the tracking of regions of cloud is likely to be complicated by the apparent differential movement of cloud at different distances. For this phase of development we have chosen to use a ground-based camera to reduce the obscuration of clouds. A stationary platform means that the relative motion will be only due to the wind.


C17

Classifying the image into clouds and bluesky is non-trivial. The disintegration of old clouds means that there is a continuum between dense cloud and blue-sky. There is no ‘correct’ dividing line and indeed we do not want one. Instead we measure how closely each image pixel resembles blue sky and how closely it resembles cloud. This appearance is purely based upon colour information.

Figure 3: Visual Picture of a Cloudscape

Identifying cloud as possibly expanding due to thermal activity requires that the cloud be tracked over an extended period of time. This tracking must be done on a sufficiently fine spatial scale that the expansion of the cloud may be distinguished from movement due to the wind. The approach adopted is based upon image patch correlation. Where the correlation value (sum of colour differences) is smallest this should correspond to the same patch. In practice the movement between images is small and only a few patch comparisons are required, but the small movement means that a subpixel flow vector is required if rounding errors are to be kept under control. An accumulated flow field is calculated by combining the results of many pair wise flow estimates. It is hoped that expanding (and contracting) clouds may be identified from where this measured flow field diverges from the flow due only to wind. Trials later this year will establish whether the approach is viable and the time frame over which expansion of clouds might be detected. The objective is to recommend regions (time bounded cylinders) to avoid or in which to seek lift. For each region a (x,y) centre will be determined from the identified bearing by converting the deduced angle of elevation of the base of the relevant cloud, knowing the actual cloudbase height. Waypoint Selection

Figure 4: Classification of Cloudscape

The cloudscape shown in Figure 3 illustrates how cloud may gradually merge with blue sky. The wispy cloud may well be associated with descending air and the large areas of grey and cyan in the processed image, Figure 4, form the basis for marking areas of the sky to avoid.

The soaring opportunity predictions detailed in the preceding sections provide the base data to allow the generation of a flight path for an AAV to exploit soaring energy within the local atmosphere. For the purposes of the Phase 2 trials the flight path is described as a number of waypoints. These waypoints can be either selected manually or autonomously. From the current sailplane GPS position a required bearing to the next waypoint is calculated. This bearing requirement is transmitted as a voice instruction transmitted from the


C17

ground station to the conventional radio link.

pilot

by

a

The rationale for providing a manual generation of waypoints function is to allow a survey of the local atmosphere to be undertaken. During the Phase 2 demonstrations these surveys will allow the accuracy of the soaring predictions to be assessed against measured data. The automated waypoint selection method is described in detail in the Agent Architecture section below. Agent Architecture The agent architecture will be able to receive input data from physical sensors and interpret that data in order to plan a detailed sequence of waypoints to be flown.

airborne vehicles and also with agents on the ground, e.g. associated with groundbased sensor platforms. The ability to modify with ease the physical distribution of agents has been employed in this project. Initially, sensors are connected directly to a UHF wireless modem for communication down to the agent colony running on a laptop computer on the ground. In a real AAV, the majority of the agents would be airborne, running on a small form-factor computer – the same software architecture would be applicable, although the necessary air/ground data rate requirements would be reduced.

The approach taken is to use a multi-agent reasoning system wherein the agents communicate via a ‘blackboard’. Conceptually, the blackboard can be regarded as a map, on which relevant information can be posted.

Agent technologies allow for the implementation of very loosely-coupled software objects. Those objects are able to communicate requests, commands and data using standardised protocols, such as those specified by the Foundation for Intelligent Physical Agents (FIPA – www.fipa.org). Such protocols provide for direct tasking of agents and also for more complex interactions, including competition between agents to provide a particular service.

The reasoning system consists of agents that cooperate. The architecture allows for those agents to be physically distributed, with agent colonies existing on multiple airborne vehicles with agent interaction taking place between colonies on those

An example of competition between agents could involve a future multi-platform mission where several simple ‘thermal hunters’ might be available for tasking by a more sophisticated asset such as a communications platform.

Figure 5: Target Agent Architecture


C17

Figure 5 illustrates the various specialised agents that are being prototyped. Their functionality is summarised below: • A blackboard agent. This agent maintains a ‘bag’ containing scenario objects with parameters such as position and confidence level (including temporal limits for that confidence – a kind of ‘best before date’) for information. Scenario objects will include position of self and any peers, areas of lift and sink, useful areas of orographic lift (such as ridges) and boundaries of restricted areas etc. The agent is more than just a data repository. It can answer questions such as “What areas of potential lift exist in the path x1, y1 to x2, y2. • A number of sailplane status monitoring agents, which interface to physical sensors on-board the glider. Whilst not all are being implemented at this stage, nevertheless the architecture allows for: • GPS location monitoring agent. This receives NMEA sentences over a serial interface from a connected GPS module. It determines current location and posts that information to the blackboard agent. • A temperature / humidity monitoring agent. This receives information over a serial interface from a sensing module. It determines current cloudbase and posts that information to the blackboard agent. • An altitude / climb / descent monitoring agent. This receives information from a connected variometer or combined GPS/pressure sensor capable of making instantaneous altitude measurements to determine current height and the rate of change. • An energetics status monitoring agent. This receives battery level information and reports available power parameters etc to the blackboard agent. • A mission payload monitoring agent. This receives information from the payload and reports relevant parameters,

including whether the payload is still capable of being tasked (e.g. digital camera memory full, battery empty) to the blackboard agent. • The orographic lift prediction agent is an agent that is responsible for locating areas of orographic lift. Met Office predicted airflow data is read in by the orographic agent and posted onto the blackboard. Initially, this data is used directly. Future extensions would allow an AAV to maintain multiple data sets, selecting the most appropriate based on current wind speed and direction. • The thermals prediction agent(s). Predicting the possible location of thermals involves at least two agents. A ground-based agent receives data from ground-based cloud-watching sensors. The sensors track clouds as they form, move across the sky and decay. The ground-based agent records such trend information and communicates that information to the blackboard on the airborne platform. An airborne thermals agent is able to reason over such stored information and can make predictions based on a variety of information including the likelihood of thermals being generated over land usage features, such as large car parks adjoining lakes or reservoirs. The airborne agent can also make use of information stored on the blackboard that was supplied by other agent colonies located on other airborne (or non-airborne) vehicles in the area. • The local knowledge agent. This is preloaded with information such as, “You always get good lift over there if the wind is from the East”. The agent reads a suitable format data file and presents the information to the blackboard agent. • The mission requirement agent understands the current mission objective – e.g. communications relay or patrolling a pipeline. It determines the


C17

volume in which the AAV must patrol whilst on station and under what circumstances the AAV should divert from station to gain lift. • An agent known as a strategic planner. This agent determines a ‘high level’ plan of what the AAV is to do next. The strategic planner decides when the glider should actively seek lift and when it should just fly from A to B. It considers the current aircraft status information and ‘global picture’ information maintained by the blackboard agent and generates a set of goals which the tactical planner agent must attempt to satisfy. It may consider various plan alternatives proposed by the tactical planner to achieve those goals and maintains a current proposed route – being an ordered list of current goals. Each goal is associated with a validity – i.e. the criteria which must be true for the goal to be applicable, including the time by which the goal must be realised. It continually monitors the applicability of the current proposed route by reviewing the applicability of the existing goals. • An agent known as a tactical planner. This agent determines how, in more detail, to execute the goals of a plan suggested by the strategic planner. Specifically, it takes a set of goals and generates a set of waypoint sequences that the aircraft can fly between to realise those goals. It proposes a number of possible plans to achieve those goals (e.g. fastest, shortest etc) and reports those to the strategic planner: each tactical plan alternative being a data structure, including a list of way points, the goal that the plan implements, plus parameters indicating the current best expectation for the goal post-conditions (expected time, location, altitude, bearing). The tactical planner will generate waypoint series that avoid areas of sinking air and danger areas or no-fly areas. It will also establish a suitable

search pattern on designated region.

reaching

the

• The waypoint manager agent generates a set of waypoints in a file suitable for reading by an external gliding support application such as WinPilot or Strepla which have strong gliding-centric user interfaces. Agent-Based Route Planning The operation of the two planning agents is illustrated by the following example scenario. First, it is supposed that the strategic planner devises a plan (expressed as a set of goals), which consists the following steps (to be executed in sequence). 1. Go to location x, y (say), which is expected to be (near) a source of thermals. 2. Search for a thermal near x, y. 3. Use the thermal to gain height. 4. Return to station In devising this high level plan, the strategic planner will use a prediction of the AAV’s behaviour in order to determine the plan’s outcome. This prediction makes assumptions about • how quickly the AAV can move to location x, y, • the height loss involved in doing so, • how long it takes actually to locate a thermal, • what is the likelihood that the thermal will still exist at the time that the AAV reaches it, • what the expected height achieved by rising with the thermal will be, etc. Once the high level plan has been determined, it is the responsibility of the tactical planner to generate a detailed set of way points that will achieve the desired set of goals.


C17

1. To go to location x, y the AAV must first be turned onto the appropriate heading. The AAV can then fly (glide) to x, y (at the speed assumed by the strategic planner). 2. To search for a thermal near x, y the AAV must initiate and execute a particular search pattern (e.g. zigzag across the wind). This will again involve turning the glider, and (perhaps) changing its speed. Note that the tactical agent does not itself fly the glider, rather it generates a detailed waypoint sequence which can then be presented to the sailplane pilot or an AAV flight management system. The tactical agent takes care of generating a waypoint plan that avoids problem areas. 3. Once a thermal has been located, a different behaviour is again indicated; the AAV must fly in a (relatively) tight circle within the thermal to gain height. (The strategic planner will have specified what target height is to be achieved). The agent colony will monitor current altitude, heading and location. The strategic planner will monitor current status against its current set of goals. If current status deviates from the strategic planner’s expectation then it will re-plan and re-task the tactical agent. 4. To return to station (if appropriate), again the AAV must first be turned onto the appropriate heading, before flying back. And again, the strategic planner will have specified at what speed to go. 5. Finally, if appropriate, the AAV is expected to remain ‘on station’. Of course, the AAV cannot hover in the air to do this, and must fly some holding pattern around the ‘on station’ position. Note that, as the individual steps of the high level plan are executed, details of subsequent steps may need to be amended. For example, if the search for a thermal takes longer than expected, the AAV will have to remain with the rising thermal

longer in order to achieve the target height. And, if the search moved the AAV from the ‘nominal’ position of the thermal, this move has to be taken into account when determining how the AAV returns to its station (potentially affecting both the bearing to fly and the time required for the flight). The tactical planner will be continuously monitoring the AAV’s status and, if necessary, can abort the current plan and replace it with another that meets the goals set by the strategic planner. Similarly, the strategic planner can abort the current highlevel plan (i.e. the series of goals that the tactical planner is attempting to achieve). A particularly important instance where this is likely to occur is if the AAV’s ground controller requests that the AAV return to base and land. Phase 2 Demonstrations Trials The demonstration trials within the Phase 2 activities can be divided into four phases. Shakedown Trials As stated previously in this paper the shakedown trials will be undertaken early in the planned schedule to demonstrate basic operation of hardware and reduce risk to the research programme. These trials will be used to demonstrate successful engagement of BGGC to get a glider and instructor into the air, the operation of the data comms links, GPS receiver and recorder and passing of directional instructions from the ground station operator to the sailplane pilot via a voice link. Atmosphere data measured from these initial flights will be used to provide some validate data for the initial Met Models. Orographic Demonstration Trials During the spring of 2007 a series of orographic soaring trials will be


C17

undertaken. There trials will demonstrate the capability of the Orographic Met Model and the waypoint generator functions to determine a flight path through the local atmosphere which exploits soaring opportunities from the likes of ridges whilst avoiding sinking air. Thermal Soaring Demonstration Trials The thermal soaring trials will take place in summer 2007 in a similar fashion to the orographic lift trial, the difference being that potential areas of thermal lift and sink will be identified by a combination of real time cloud analysis and predictions based on land-type and other parameters.

make full exploitation of the soaring opportunities within the local atmosphere. References [1] Vosper, S, “Development and Testing of a High Resolution Mountain-Wave Forecasting System”, 2002, Meteorol. Appl. 10, 75-86 [2] Vosper, S.B. & Mobbs, S.D, “Lee Waves over the English Lake District”, 1996, Q. J. R. Meteorol. Soc. 122, 1283-1305.

Acknowledgements The work reported in this paper was funded by the Systems Engineering for Autonomous Systems (SEAS) Defence Technology Centre established by the UK Ministry of Defence.

Combined Soaring Trials If the orographic and thermal soaring trials are sufficiently successful a final demonstration is planned that will incorporate both orographic and thermal soaring opportunities. This trial will also feature more sophisticated route planning with ‘realistic’ mission constraints. Further Developments Assuming that Phase 2 of the Autonomous Soaring Projects is successful then Phase 3 is likely to contain a number of further developments. In addition to orographic ridge lift and thermal soaring the use of other mechanisms of soaring within the local atmosphere may be demonstrated. For example Lee waves and dynamic soaring opportunities could be predicted and their exploitation demonstrated. The use of a fully autonomous AAV rather than a surrogate glider could be established. This would involve replacement of the manned sailplane with an autonomous sailplane. Current thinking suggests the engagement of the BAE Systems HERTI AAV whose airframe is based on a Polish manned glider design and as such could


C17