Dspace: Real-time 3D Visualization System For Spacecraft .

SMC-IT #19The GraphicsObject class imports scene geometry inVRML format, supports multiple levels-of-detail on a perGraphics Object basis and is used to represent both physicalobjects (vehicles/terrain/planetary bodies) and ornamentalobjects (goal markers, coordinate axes, 3D text).An individual GraphicsObject’s scene graph fragment iscombined with other GraphicsObject’s scene graphfragments in the SceneFragment to form a single morecomplex scene graph. A GraphicsObject instance must beadded to a SceneFragment instance in order to be processedduring scene graph render traversals.The GraphicsObject class is also the base class for anyuser written, specialized class that requires scene graphtraversal at render time.The DspaceTerrain libraryspecializes this class for rendering terrain geometry.2) SceneFragmentIndividual scene graph fragments, contained inGraphicsObject instances, are collected by theSceneFragment class and can be organized based on type orgroup information. Physical and ornamental GraphicsObectscan be segregated for controlled display during syntheticcamera imaging or traversal planning. For example, whengenerating synthetic camera imagery, SceneFragmentscontaining ornamental GraphicsObjects can be directed to“hide” themselves during rendering, thereby causing finalrender images to contain only physical objects such as terrainor rovers.Through the use of this mechanism,SceneFragments can be shown or hidden as directed by thesimulation as a way to selectively render collections ofGraphicsObjects at render time.3) SceneFragmentManagerPrior to the rendering cycle traversal, scene graphs arecollected into a more complex scene graph and contained inthe SceneFragment. Multiple SceneFragments are thenpassed to the SceneFragmentManager, which makes thecontained scene graph available to the SceneVantage alongwith functions for SceneFragment management such as theaddition of new SceneFragments and the deletion ofcontained SceneFragments.4) SceneVantageSceneFragmentManagers contain scene graph fragmentsand Viewports manage scene graph traversal at render time,but something must connect the scene graph to the Viewport.This connecting code resides in the SceneVantage class. TheSceneVantage class contains a scene graph as provided by aSceneFragmentManager (the “Scene”) and a camera or set ofcameras (the “Vantage”). When directed by the Viewport atrender time, the SceneVantage instances traverse their entirecontained scene graphs and viewing the scene geometryusing the selected Camera. Fig. 3 shows how multipleSceneFragments and SceneFragmentManagers are utilizedby one or more viewports.Figure 3. GraphicsObjects, SceneFragmentManagers and Viewportorganization allow segregated rendering.SceneVantage instances can contain multiple Camerasbut only one selected Camera may be active at any giventime. Scene graph traversal is typical for a retained moderenderer, but if shadows are activated, a two pass renderingprocess based on the Z-buffer shadowing [11] algorithm isperformed, with final shadow determination performed in afragment shader.5) FramesFrame classes provide the backbone or skeletal systemfor the entire scene graph and provide a mechanism forarticulating the components of multi-body objects, such asroving vehicles.Frame instances provide four important functions atruntime. First, Frames provide an attachment point forgeometry as shown in Fig. 4. For example, a rover chassisframe may have the geometry for the chassis, solar andcamera mast all attached. Second, Frames can be attached toother Frames, thereby forming the parent-child hierarchicalscene graph. Third, transformation information as receivedfrom the Dshell simulation is stored in Frame instances.When transformations are applied to a Frame, all attachedgeometry are translated, rotated or scaled accordingly.Fourth, when specified as Physical or Ornamental, attachedGraphicsObjects with their contained geometry are providedto the appropriate SceneFragment by the Frame, therebysegregating physical from ornamental objects at render time.This is important when performing selective rendering, asdiscussed later and shown in Fig. 7, for synthetic camera

SMC-IT #19III.DSHELL SIMULATION-DSPACE INTEGRATIONDuring a Dshell simulation run, the initializationprocess is responsible for registering simulation objects withDspace in the correct scene graph hierarchy. Dshell doesthis by directing Dspace to create Dspace Frame instances,which in essence are proxy classes for Dshell ’s ownFrame instances. Every Dshell simulation object, fromplanetary bodies to individual parts of a multi-body system,such as a spacecraft or roving vehicle, contain a Dshell Frame.A. Synchronizing the simulation and the visualizationFigure 4. Dspace frames provide scene graph backbone.imagery where only physical objects (terrain, vehicles,planets) are to be rendered.6) penGL cameras: Perspective, Orthographic andFrustum. These cameras can be arbitrarily placed andoriented under user control, either through mouse interaction,or through simulation control via the Dspace API. TypicalOpenGL camera operations are supported such as position,orientation, frustum settings and clipping plane values. TheDspace camera classes are derived from the Coin3D baseclasses and provide convenience routines for positioning andpointing cameras when following designated simulationobjects. This mechanism is used to provide “chase” views ofspacecraft or rover vehicles.7) ViewportsThe Dspace Viewport class is a container for a singleScene Vantage instance and also provides routines used forperforming scene graph traversal and rendering, userinteraction and event handling (mouse or keyboard),interaction with the native windowing system (typicallyX11) and displaying the final rendered 3D scene imagesalong with any simulation provided tabular engineeringinformation. Users can create, resize, hide and placeViewports, control camera position and pointing and select3D scene items via mouse interaction. Dspace allows usersto create an arbitrary number of Viewports, only limited byworkstation resources.As simulation object status is updated during simulationexecution, Dspace’s scene must be updated accordingly. TheDshell simulation manages this efficiently at run-timeusing watch handlers that trigger only when the state of asimulation object changes. For articulated rover vehiclesimulations, joint motions for vehicle components, such assuspension systems or camera mast parts, trigger simulationcallbacks that send frame information (position andorientation) to the corresponding Dspace Frame. UpdatingDspace Frame information happens continuously in real timeand keeps the Dspace visualization state in sync with thesimulation state at all times while minimizing the amount ofupdate traffic flowing from the simulation to Dspace.B. Kinematic AccuracyBecause users rely significantly on the Dspacevisualization to interpret the simulation state, it is essentialthat the visualization model correspond accurately to thesimulation model geometry and kinematics.Keeping graphics models consistent with simulationmodels can be a challenging and expensive proposition asthe vehicle structure and design evolves, especially duringthe early phases of projects. The described one-to-onecorrespondence between Dshell Frames and Dspace proxyFrames mitigates this potential problem. This approachessentially consists of Dshell auto-generating a skeletonscene graph that is derived from, and faithful to theunderlying physical model within the simulation. The 3Dparts geometry of the vehicle, such as wheels, masts, arms,etc., are attached to this skeleton to ensure that the simulationand visualization models maintain close correspondence toeach other. In this way, Dshell essentially directs Dspaceto construct the 3D geometry of the scene based on accuratesimulation data each time that the simulation and Dspace runas in Fig. 5. If a vehicle component, specification orcharacteristic changes from run to run, only simulationchanges are needed, no changes to Dspace are required.

SMC-IT #19terrain, inertial and body fixed coordinates are returnedthat can be used by the Dshell simulation to set vehiclegoal information or vehicle placement.1) Rich User APIFigure 5. Dshell to Dspace Frame connections ensure kinematicaccuracy.IV.When running in single process (local) mode, withDspace linked as a library, all of Dspace’s provided coreclasses, as described above, can be invoked via C orPython. We use SWIG [14] to generate the Python to C interface classes.When running in distributed (server) mode, with Dspaceexecuting as a separate process from the simulation, weprovide Python proxy classes for the core Dspace classes.The Python proxy classes send Python commands to Dspacevia Python sockets. Because Dspace is a library of Pythonwrapped C classes, the user is free to create a Dspaceapplication using either a C main routine or a Pythonscript. When running as a Python application, users canaccess Dspace class instances and invoke class memberfunctions.When running in either local or server modes, users cancustom-craft the look and feel of their Dspace visualization.DSPACE FEATURE SETA. High Performance RenderingTo support the increasing complexity of space missionsimulations, Dspace combines fixed functionality OpenGLwith OpenGL Shading Language [12] (GLSL) shader codeto perform rendering operations that would be prohibitivelyslow with fixed functionality OpenGL alone. GLSL shadercode essentially re-programs the graphics processing unit(GPU), on the workstation’s graphics card, to performcustom operations that may not be supported using fixedfunctionality OpenGL, and executes that shader codeconcurrently on the GPU’s multiple processing cores. Asdescribed above, vehicle shadows, performed in a two passrendering process and computed in a fragment shader can berendered, while maintaining 30 fps frame rates. Fig. 6shows a Dspace rendering with rover shadows. Vehiclewheel tracks, terrain characteristic information such as heightand normal maps, multi-texturing and the use of the Perlin[13] noise algorithm to enhance terrain realism are alsoperformed in OpenGL Shading Language vertex andfragment shaders.B. User InteractionUsers can interact with the Dspace rendered scene andcan modify camera viewpoint information via mouseinteraction. In addition, Dspace can be used to acquire dataabout the 3D scene. For example, by picking a point on theFigure 6. Rover camera view with real-time shadows2) Selective RenderingRendered scenes typically contain "real" simulationobjects together with additional "annotation" objects that canhelp the user interpret the state of the simulation. TheDspace framework provides ways to arbitrarily organize andgroup simulation objects by such properties to allow runtime selection of the visualization content by selectiverendering of the groups. An important use case for selectiverendering is that of rover camera modeling as describedabove, where only physical-based objects such as terrain,planetary bodies or rover components, should be seen in the

SMC-IT #19final rendered image, and ornamental objects such as goalmarkers, viewing frustum pyramids or coordinate axes are tobe excluded. Fig 7 depicts a typical scene graph backbonecreated through the use of Frame instances, with each Frameinstance referencing attached geometry, with the attachedgeometry assigned to either physical or ornamentalS

visualization source code from one simulation to another. Additional capabilities such as 30 frame per second rendering performance, realistic terrain rendering, synthetic image synthesis for vehicle camera modeling, real time vehicle shadows and wheel tracks (as shown in Fig. 1), managing physical objects such as terrain and vehicles and