Planet Surface Simulation with PANGU

Transcription

1 Planet Surface Simulation with PANGU S.M. Parkes, I. Martin, M. Dunstan, D. Matthews University of Dundee, Dundee, DD 4HN, Scotland, UK. Planetary landers have, in the past, relied on physical means to protect the payload from the shock of impact on the surface []. These landers, starting their descent from orbit with their initial position only known to a few kilometres, were not required to land at a particular landing spot, but only to land safely. Today, much more knowledge, obtained from earlier landings and high-resolution orbiting instruments, is available about the surfaces of some planets than was available when previous landers were designed. Missions are becoming more demanding in terms of the accuracy of landing and significant effort is now focused on the design of surface relative navigation systems. Surface relative navigation requires a sensor that can pick out features or landmarks on the surface and use these to track the position of the spacecraft relative to the surface passive and active vision-based navigation sensors are currently being developed. The testing of these sophisticated sensors, in particular the image processing parts, required the development of a realistic, large-scale test bed, representative of the real planet s surface. Physical modelling was not able to meet the needs of the sensor testing, so a virtual reality tool has been developed. PANGU (Planet and Asteroid Natural Scene Generation Utility) is a software tool for simulating and visualising the surface of various planetary bodies. It has been designed to support the development of planetary landers that use computer vision to navigate towards the surface and to avoid any obstacles near the landing site. PANGU can be used to generate an artificial surface representative of cratered planets and to provide images of the simulated planet. When given the position and orientation of a spacecraft above the planet s surface, PANGU responds by producing an image of the surface from that view point. Current research is extending the capabilities of PANGU so that Martian surfaces and asteroids can also be simulated. This paper describes the PANGU simulation tool in detail and provides example images of the simulated surface as seen from a descending planetary lander. of

2 PANGU Overview The PANGU (Planet and Asteroid Natural scene Generation Utility) tool was originally developed specifically to simulate the surface of heavily cratered planetary bodies like the Moon and Mercury. It is currently being developed to simulate Martian surfaces and asteroids. PANGU first builds an initial surface model on to which it adds appropriate surface features like craters and boulders. Initial Surface Model The initial surface model may be entirely synthetic, constructed using fractal techniques, or it may be based on an existing digital elevation model (DEM) e.g. MOLA (Mars Orbiting Laser Altimeter) DEM. When a DEM exists for a planet surface then this can be used to give a real coarse resolution surface model. Of course, if a sufficiently high resolution DEM is available there is no need to use PANGU facilities to build a simulated surface. If the existing DEM is of low resolution (e.g. MOLA data is around 500 m resolution horizontally) then synthetic, but realistic, detail has to be added to it. PANGU can do this using fractal interpolation techniques to add surface detail to the real low resolution DEM. Appropriate small scale surface features can also be added like small craters and boulders. Crater Simulation PANGU builds a cratered planet surface model by simulating impact cratering on an initial terrain model [2], [3], [4], [5], [6], [7], [8] Craters are placed on the terrain either manually or randomly according to a user defined crater size-density distribution. The crater models combine idealised mathematical impact crater models [9], [0], [] with fractal techniques to produce a realistic appearance to the craters. The crater bowl, rim and ejecta blanket are simulated. Craters are caused by the high velocity impact of meteorites. Large amounts of energy are released during impact [9]. There are four basic types of crater that appear on the lunar surface. Simple Craters up to 0-20 km diameter which have a bowl shaped appearance. Complex Craters above 0-20 km diameter which have steep walls and relatively flat floors. Features observed within this type of crater include terraced walls and central peaks. Multi-Ringed Craters over several hundred kilometers in diameter which take the form of several concentric rings. Secondary Craters small craters caused by the ejecta from primary craters impacting the surface. They have a more rounded appearance than the simple crater due to the lower impact velocity. Similar types of crater appear on Mercury. On Mars the craters are generally similar but are affected significantly by Aeolian processes. 2 of 2

3 A basic simple-crater model is illustrated in Fig.. It comprises the almost parabolic crater bowl which penetrates the original surface, the crater rim which protrudes above the original surface and the ejecta blanket. R im H e ig h t, H r C rater H eight, H Ejecta Blanket O rig in a l S u rfa ce C r a te r B o w l C rater D iam eter, D Figure : Basic Crater Model A simulated crater produced with the basic crater model will not look very realistic. To improve its appearance fractal zones may be added to the basic model to give the crater s surface appropriate texture. Fig. 2 shows two fractal zones being added to the basic crater model. Basic Crater Model Floor Wall Basic Ejecta Model Fractal Zones Floor Wall Underlying Terrain Fractal Ejecta Fractal Zones added to Basic Crater Model Final Ejecta Surface Figure 2: Crater Bowl Model Figure 3: Ejecta Blanket Model When a meteorite impacts at high velocity the original surface is obliterated leaving the crater bowl. Shock waves from the impact force the surrounding terrain to rise close to the outer edge of the crater. Material ejected from the crater lands on the surrounding original surface. This may be modelled as shown in Fig. 3. The basic ejecta model takes into account the effects of the shock wave and the average accumulation of debris close to the crater. The form of the ejecta material is modelled using fractals and added with the basic ejecta model to the underlying terrain. An important feature of craters is their size-density distribution. There are usually many more small craters than large ones. To make a simulated crater planetary surface look realistic this must be taken into account. The actual crater size-density distributions for the Moon, Mercury and Mars vary across their surfaces. 3 of 3

4 A fresh simple crater has sharp rims and a clearly defined bowl-like interior with rough ejecta spreading outwards from the rim. Crater degradation occurs over long time periods due to the continuing effects of later impact craters - from the smoothing effect of micro-craters to the settlement of material due to large impacts and from aeolian processes on Mars. Over a large period of time crater features become less sharp, rims more rounded and the bowl less deep. Eventually all that is left is a depression that is itself, eventually completely levelled. The degradation of an individual crater is dependent on its diameter and on the amount of impact bombardment that the area has undergone since the crater formed. There is little recent published material on crater degradation but Ross in 968 [0] and Soderblom in 970 [] created separate theoretical models of crater degradation based on topographic displacement caused by small meteor impacts. The rate of crater erosion is related to diameter with smaller craters eroding faster. Crater erosional characteristics are specified in PANGU using two separate graphs which relate normalised crater depth and rim height to age for a crater of a specified diameter. This is defined as the baseline crater diameter. Examples of PANGU baseline crater rim and depth erosion graphs are shown in Fig. 4. Normalised Crater Depth Time (billion years) Diameter 00 Density Power Normalised Rim Height Time (billion years) Diameter 00 Density Power Figure 4: Graphs Showing Effect of Erosion on Crater Depth and Rim Height To add random craters to a surface, crater diameters are specified using a crater size distribution. Each crater is also assigned an age from a specified crater age distribution, see Fig. 5. The baseline age and hence the erosional state of the crater is calculated from the crater age taking into account the crater diameter. Fig. 6 shows nine identically sized craters of varying age and erosional state from fresh though to completely obliterated. 4 of 4

5 Relative Crater Age Probability Distribution.0E+00.0E-0.0E-02.0E-03.0E-04.0E-05.0E Crater age in billion years Relative Probability Figure 5: Crater Age Distribution Figure 6: Craters of Various Erosional States Other Features Other surface features that can be simulated in PANGU include boulders and sand dunes. Boulders are simulated as complete semi-spherical faceted objects which are sized according to a boulder size-density distribution and placed into the surface. Boulders are embedded into the surface a variable percentage of their size to add to the realism of the models. Boulder position can be made a function of position relative to a crater centre, to help simulate boulders in an ejecta blanket. Sand dunes are simulated using a simple barchan dune model. Other forms of sand dune are constructed by joining multiple barchan dunes. This enables two principal parameters to be used to specify a range of different dune formations: sand quantity and predominant wind direction. When there is little sand barchan dunes will form. When there is substantial sand these dunes will join together to form ridge dunes. All PANGU models and their distribution on the surface model are parameterised using parameters familiar to planetary scientists to simplify the generation of terrains similar to those expected and characterised by mission scientists. PANGU System Architecture The structure of the PANGU software is illustrated in Fig. 7, rectangles represent software programs and cylinders represent files. The Surface Generator is responsible for producing the simulated planetary surface. It takes in various parameters from files and uses this information to build the required surface. The Surface Parameters file specifies the parameters that the Surface Generator uses when creating an initial surface model. Surface features are added to the initial surface model according to the information held in the feature lists and feature model parameters files. There is one feature list for each type of feature to be added to 5 of 5

6 the surface. It contains the position, size, age, etc. of each crater, boulder or other surface feature. The generic form of each type of feature is held in a feature model file. Again there is one feature model file for each type of feature being added to the surface. The surface generator initially forms a DEM which is subsequently converted into a surface polygon format. All the surface features except boulders are added to the DEM. Boulders are added to the Surface Polygon file because they cannot be adequately represented in a DEM. Surface Parameters Feature Lists Feature Models Sensor Model Parameters Illum inate Parameters Surface Generator Surface Poly gons Im ag e Generator Sensor Coordinates Sensor Image TCP/IP Socket Inter fa ce Surface DEM Shadow Cast Shadow Map Figure 7: PANGU Architecture The Surface Polygons file forms the interface between the Surface Generator and the Image Generator. The Image Generator takes the surface model from the Surface Polygons file and renders it in a way appropriate to the type of sensor being simulated to produce images of the surface. A camera simulation requires information about the camera and illumination sources, which are stored in a Camera Model Parameters and Illuminate Parameters files. For visual simulation shadows have to be cast across the surface depending upon the relative position of the light source (Sun). Shadow casting is done off-line as it is a fairly computationally intensive task. For LIDAR simulation no shadow casting is necessary. The relative position and orientation of the sensor is specified through a TCP/IP socket interface. The Image generator takes the camera co-ordinates and renders an image of the surface from that position. The image is then returned to the TCP/IP socket. PANGU is easy to connect to another system, e.g. an image processing system or GNC simulation. The TCP/IP socket interface to PANGU enables any Internet enabled computer to connect to the PANGU tool, control it and receive images from the surface simulation. The Image Generator component of PANGU acts as a server, listening for a connection to be established with another machine. A client running on the other system opens a TCP/IP socket connection with the PANGU server and transmits the sensor coordinates over the connected socket. The PANGU server then generates the image and sends it back over the socket connection to the client system. 6 of 6

7 Visual Images Fig. 8 shows an example image of a heavily cratered surface produced by PANGU. Fig. 9 shows a Martian surface with a large number of boulders similar to the Viking and Mars Pathfinder landing sites. Figure 8: Heavily Cratered Surface Figure 9: Martian Surface 7 of 7

8 Scanning LIDAR Simulation As well as producing visual images PANGU can produce images from a scanning LIDAR. A scanning LIDAR is illustrated in Fig. 0. A laser pulse is transmitted from the spacecraft and propagates towards the surface, spreading a little as it does so. When it hits the surface some of the energy is reflected back towards the spacecraft and is received by the LIDAR receiver. The time-of-flight of the pulse gives the range to the surface. By scanning the LIDAR beam and emitting a sequence of pulses at each beam position a three dimensional model of the surface can be built up. The PANGU tool uses ray-tracing to simulate each beam of the scanning LIDAR. The LIDAR sensor position and orientation is determined for each separate beam in the scan, since the scanning mechanism is relatively slow and the spacecraft will move a significant distance in the time between pulses. A single beam is then projected in the appropriate direction and its interception with the surface model determined. The distance from the LIDAR to the surface is calculated along with the slope of the surface relative to the beam. This information is used to form one pixel of the LIDAR range image. The complete LIDAR image is formed as the beam is scanned in the x and y directions by the LIDAR instrument. If necessary multiple beams can be used to form a single LIDAR pixel to prevent aliasing. Range Beam Divergence Single Beam Scanning Beam Figure 0: LIDAR Range and Scanning In Fig. an image of a synthetic dune covered terrain is shown together with a colour coded LIDAR depth image, where the LIDAR image is taken from the same position as the visual image. 8 of 8

9 Figure : Visual and LIDAR Images of Dunes. Conclusions PANGU is a powerful set of tools for supporting the development of autonomous, actively guided, planetary landers. Simulated surface models of several different planetary bodies can be produced based on parametric information from planetary scientists. If a low resolution digital elevation model is available for a particular planetary surface it may be used as the starting point for a PANGU simulation with the PANGU tools being used to add detail and small scale surface features to the surface model. Cameras and LIDAR instruments can currently be simulated within PANGU. Image generation is fairly fast allowing near real-time simulation for the testing of image processing and guidance and navigation control algorithms. As well as sensor simulation for planetary landers PANGU may be used for rover navigation, and the testing of other forms of sensor like planet limb sensors and asteroid trackers. PANGU is currently being used to support the development of a vision-based navigation camera and a LIDAR hazard detection system for planetary landers. Acknowledgements The Authors would like to acknowledge the support of ESA for the LunarSim study (ESA Contract No. 282/98/NL/MV), the PANGU study (ESA Contract No. 747/95/NL/JG) and the Asteroid and Whole Planet Simulation with PANGU study (ESA Contract No. 7338/03/NL/LvH/bj) on which this paper is based. The assistance of Salvatore Mancuso, Patrick Plancke, Stein Strandmoe, and Giovanni Bolognese, the ESA study technical managers for these and related studies, is much appreciated. 9 of 9

10 References [] T.D. Halbrook, J.D. Chapel and J.J. Witte, Derivation of Hazard Sensing and Avoidance Maneuver Requirements for Planetary Landers, Guidance and Control, Advances in the Astronautical Sciences, Volume 07, American Astronautical Society, (200). [2] S.M. Parkes, I. Martin and I. Milne, Lunar Surface Simulation Modelling for Vision Guided Lunar Landers, Proc. DASIA 99, ISBN (999). [3] S.M. Parkes and I. Martin, Virtual Lunar Landscapes for Testing Vision-Guided Lunar Landers, Proc. IEEE Int. Conf. on Information Visualisation (IV 99), IEEE Computer Society, ISBN (999) [4] S.M. Parkes and I. Martin, Virtual Reality for Real Spaceships, Proc. International Conference on Virtual Systems and MultiMedia (VSMM 99), -3 September 999, Dundee, Scotland, pp 56-68, ISBN (999). [5] S.M. Parkes and I. Martin, Lunar Surface Simulation Opening the Road to the Moon, Proc. Fourth International Conference on the Exploration and Utilisation of the Moon, Noordwijk, The Netherlands, ESA Pub. No. SP-462, ISBN (2000). [6] S.M. Parkes, I. Martin and S. Strandmoe, Planet Surface Simulation for BEPI Colombo, Proc. DASIA 200, ESA Pub. No. SP-483, ISBN , (200). [7] S. M. Parkes, I. Martin, M. Dunstan and S. Mills, Mercury surface simulation for Bepi Colombo Lander, Proc. DASIA 2002 ESA Pub. No. SP-509, (2002). [8] S. M. Parkes, M. Dunstan, D. Matthews, I.Martin and V. Silva, LIDAR-based GNC for Planetary Landing: Simulation with PANGU, Proc. DASIA 2003, ESA Pub. No. 532, ISBN (2003). [9] H. J. Melosh, Impact Cratering A Geologic Process, ISBN , (Oxford University Press, 989) [0] H.P. Ross, A Simplified Mathematical Model for Lunar Crater Erosion, Journal of Geophysical Research, Vol 73, No.4, pp (968). [] L. Soderblom, A Model for Small-Impact Erosion Applied to the Lunar Surface, Journal of Geophysical Research, Vol 75, No.4, pp (970). 0 of 0