Installing the Sensable Plugin
- Download Geomagic drivers and SDK:
OpenHaptics for Windows Developer Edition v3.4
OpenHaptics for Linux Developer Edition v3.4
- The OpenHaptics SDK comes with both the Phantom Drivers and the OpenHaptics libraries. First install the drivers, then the SDK.
- Run the Phantom Test program that came with the SDK, and use it to check that the Omni is working. It is also recommended that you use it to calibrate your Omni.
- In your Sofa CMake configuration, select SOFA-PLUGIN_SENSABLE and configure. If you installed OpenHaptics in the default location, CMake should find all the libraries and directories automatically.
If CMake doesn’t find them for you, it will return an error and you can set them manually. The HD and HL libraries are found in OpenHaptics/Academic/3.1/lib/“your system’s subdirectories”, while the HDU library is found in OpenHaptics/Academic/3.1/utilities/lib/“your system’s subdirectories”. Once all the variables are set, you can Configure, Generate and compile as usual.
Using the Plugin
One method of using the Omni in a scene is to control a rigid object directly with the Omni, and compute the force feedback based on the penetration of that rigid object with any other object in the scene. The Senable Plugin contains two important components. The NewOmniDriver interfaces with the Omni device, getting the tool’s location and sending it the appropriate force feedback. The EnslavementForceFeedback computes the force feedback values, based on the collision detection that Sofa already does in a scene.
Example Scene 1 – Using the DistanceGrid Collision Model
In the examples directory of the Sensable Plugin, there are a number of example scenes. We will examine SimpleBox-DistanceGrid.scn. This scene has two main objects. A long curved tool is controlled by the Omni, and a simple box serves as something for us to feel. Taking a look at the scene in more detail, we see our collision pipeline:
This is what does the collision detection for the scene. Next, we see our NewOmniDriver:
We will discuss all the attributes in more detail later, but the important one to note now is the tags attribute. The NewOmniDriver needs to find the MechanicalObject that it will control, and the ForceFeedback that will calculate the feedback for it. It will do this by looking for the components that have the same tag as in, in this case Omni. Next we see the Instrument node. First we see the MechancialObject:
Here, the template type is important. The NewOmniDriver is looking for a MechanicalObject with the template Rigid. Also, we see that it has the matching tag, so that the NewOmniDriver knows that this is the MechanicalObject that it should be controlling. The UniformMass object simply gives the MechanicalObject some mass. In the VisualModel node, the OglModel loads the mesh that is used to visualize the instrument.
<OglModel template="ExtVec3f" name="InstrumentVisualModel" fileMesh="data/mesh/dental_instrument.obj" scale3d="10 10 10" translation="-2.12256 1.32361 35.5" rotation="180 0 150" material="Default Diffuse 1 1 0.2 0.2 1 Ambient 1 0.2 0.04 0.04 1 Specular 0 1 0.2 0.2 1 Emissive 0 1 0.2 0.2 1 Shininess 0 45" />
Note the translation and rotation attributes. We will come back to them later. The material attribute just gives the instrument a different look than the box. The RigidMapping connects the MechanicalObject we saw earlier with the OglModel. This keeps the visualization of the instrument in sync with the movement of the Omni. In the Collision Model node, there are a number of important details.
<MechanicalObject template="Vec3d" name="Particle" position="0 0 0" /> <Point name="ParticleModel" contactStiffness="2" /> <RigidMapping template="MechanicalMapping<Rigid,Vec3d>" name="MM->CM mapping" object1="instrumentState" object2="Particles" /> <EnslavementForceFeedback name="forcefeedback" tags="Omni" collisionModel1="@ParticleModel" collisionModel2="" relativeStiffness="4" attractionDistance="0.3" normalsPointOut="false"/>
First, we see the MechanicalObject named Particle, and the attribute position with the value “0 0 0”. Here we are making defining a single point that will be used for our collisions. This point represents the tip of the Omni. We place it at “0 0 0” because this lines it up with the Omni properly. We then give our particle a Point collision model. The RigidMapping, like the earlier one, keeps are particle in sync with the movement of the Omni. The visual model is also in sync with the motion, and we want to tip of the instrument to correspond with our particle, so that the tip of the instrument is where the collision occurs. That is what the translation and rotation attribute in InstrumentVisualModel are for. They line the visual model of the tool up with the particle. We put the translation in the visual model instead of in the particle so that the tip lines up with the movement and rotation of the Omni. The EnslavementForceFeedback listens for any contact that our Particle Model makes with any other model in the scene, then calculates the force feedback for the Omni accordingly. We will look into all the attributes in detail later, but for now notice that again the tag attribute matches that of the NewOmniDriver. The Box node also has a visual model and a collision model. In this case, the same mesh is used in both models. Here, the DistanceGrid is the collision model for the box.
Example Scene 2 – Using the Triangle Collision Model
The example scene SimpleBox-TriangleModel.scn sets up the same scene as SimpleBox-DistanceGrid.scn, but using a different collision model type. There are two key differences in this example:
The DistanceGrid component is replaced by
In the component MinProximityIntersection has the the attribute useSurfaceNormals set to true
The useSurfaceNormals attribute tells the MinProximityIntersection to use normals from the Triangle collision model when creating the collision information, which is then used by the EnslavementForceFeedback to correctly compute the force feedback.
Another way to set up the scene is to indirectly control a rigid object, by attaching it to the position of the Omni using a spring.
Example 3 Using a Spring
The example scene SimpleBox-Method2.scn sets up the same type of scene as the above examples, but using a spring to link the rigid body and the Omni instead of directly controlling it. The Collision Pipeline is slightly different:
We first create the representation of the Omni’s actual position. In contains the NewOmniDriver. Just like in Method 1, we use the tags attribute to match up the NewOmniDriver, the mechanical object, and the ForceFeedback (which will show up later).
<Node name="OmniObject"> <Node name="RigidLayer"> <MechanicalObject name="ToolRealPosition" tags="Omni" template="Rigid" /> <NewOmniDriver name="omniDriver1" tags="Omni" scale="300" permanent="true" listening="true"/> <Node name="Tool1"> <MechanicalObject template="Rigid" name="RealPosition"/> <SubsetMapping indices="0"/> </Node> </Node> </Node>
Then we create our actual tool that will interact with the scene. First we need solvers to allow for collision detection:
Next we create our rigid object:
And we provide a ForceFeedback component to calculate the force feedback to the Omni:
Next we create the Collision Model, and map it to the rigid model we created above:
Now we need a Visual Model, also mapped to the rigid model we create above:
<Node name="ToolVisual"> <OglModel template="ExtVec3f" name="VisualModel" fileMesh="data/mesh/dental_instrument.obj" scale3d="10 10 10" translation="-2.12256 1.32361 35.5" rotation="180 0 150" /> <RigidMapping template="Mapping<Rigid,ExtVec3f>" name="MM->VM mapping" object1="ms" object2="VisualModel" /> </Node>
Finally, we add the components to link the tool we created with the representation of the Omni’s position:
The rest of the scene contains the box, which is created just as it was in the other scenes.
- listening – When true, the NewOmniDriver will listen to the ForceFeedback for the computation of the feedback.
- tags – The NewOmniDriver will look for a MechanicalObject with template Rigid with a matching tag, and a ForceFeedback with a matching tag.
- forceScale – scales the force feedback given to the Omni. This attribute will involve a balance in your scene. If it is too high, you will feel a lot of oscillations and vibrations. Too low and you will be able to push through your surfaces easily.
- scale – scales the motion from the Omni. This changes how much motion on the physical Omni it takes to move the tool in the scene.
- permanent – True if the force feedback will be applied permanently.
- alignOmniWithCamera – True if the object controlled by the Omni, and the direction of motion from the Omni, should remain lined up with the camera. This means that as you change the camera’s view, moving the Omni arm up will always make your object move up in the view. By default, the scene’s default Camera is used, but if you wish to specify a camera, you can do so by matching the camera’s tags with the NewOmniDriver’s tags. See Example scene CameraAlignment.scn and SpecifyingCamera.scn.
- tags – should correspond with the tags attribute in the NewOmniDriver
- collisionModel1 – The collision model that the Omni controls. The EnslavementForceFeedback will then gather the collision information for every collision this model has in the scene
- collisionModel2 – If you are only interested in the collision between collisionModel1 and one other collision model in the scene, you can specify that other model here. The EnslavementForceFeedback will then only gather the collision information for collisions involving both models. Note: if you do not want to specify another model, it is important that you specify the attribute as collisionModel2="".
- relativeStiffness – To reduce oscillations when contact is made with a surface, the force applied when the instrument is found to be inside the other object is higher than when it is outside. The ratio of these forces is specified by the relativeStiffness. This value will also involve a balance in your scene. If it is too high, you will feel a lot of oscillations and vibrations. Too low and you will be able to push through your surfaces easily.
- attractionDistance – To reduce oscillation when contact is made with a surface, as the instrument gets very close to making contact, it is slightly attracted to the contact point. Once it reaches the contact point, the force feedback will push away from the object, to prevent the instrument from entering. The distance from contact at which this attraction starts is given with attractionDistance.
- normalsPointOut – In order for the force feedback to be applied in the appropriate direction, the EnslavementForceField must know if the normals of the other objects in the scene point towards the inside of the object or towards the outside. If they point towards the outside, normalsPointOut should be true, and false in the normals point towards the inside. Note: All the objects (other than the once being controlled by the Omni) need to have their normals pointing in the same direction.
- contactScale – This scales the strength of the force feedback depending on the size of the objects in your scene. Larger objects will require a larger contactScale in order for the force feedback to be felt.
Scenes at different scales
The forces computed in your scene will vary depending on the scale of the objects in your scene. For example, the box in SimpleBox.scn is 10x10x10, while the box in SimpleBoxLarge.scn is scaled to be 10 times bigger than that. At larger and smaller scales, a number of your attribuets will need to become larger and smaller as well. The attributes that are sensitive to scale are:
- MinProximityIntersection: alarmDistance
- MinProximityIntersection: contactDistance
- NewOmniDriver: scale
- EnslavementForceFeedback: attractionDistance
- EnslavementForceFeedback: contactScale
Look at the differences between these attributes in SimpleBox.scn and SimpleBoxLarge.scn to learn more.
Scenes with translated objects
One of the downsides of the DistanceGrid collision models is that it doesn’t allow the EnslavementForceFeedback to correctly compute the force feedback on an mesh that has been translated in the scene. There are two ways to deal with this:
- Use software such as Blender to create a new mesh that has the translation built into the mesh file. See example scene TwoTeeth-DistanceGrid.scn
- Use the Triangle collision model instead of the DistanceGrid. See example scene TwoTeeth-TriangleModel.scn
Rotating the scene
By default, when you rotate a scene with the NewOmniDriver, the tool being controlled by the Omni follows this rotation. In this way, no matter how you are viewing the scene, moving the Omni arm up will move the tool up in the view. This behavior can be turned off by setting the attribute alignOmniWithCamera to false. See the example scene CameraAlignment.scn. The tool aligns to the view of a specific camera. Usually, a scene won’t have a camera specified, and the NewOmniDriver will align the tool to the default view. If you do have a camera specified, it will align with that camera. If you have more than one camera specified, you can specify the camera that the tool will be aligned to by setting the tags attribute to match the tags of the NewOmniDriver. See the example scene SpecifyingCamera.scn.
Last modified: 14 April 2020