This document provides an overview of the DIRSIG5 JSON formatted simulation description, referred to as JSIM.
Introduction
DIRSIG5 utilizes a suite of plugin modules to assemble the desired simulation capabilities at execution time. The initial set of plugins were written to establish backward compatibility with DIRSIG4 atmospheric and sensor descriptions. As DIRSIG5 has matured, more features have been added to the overall model through plugins.
Although JSIM is a powerful tool for configuring DIRSIG5 simulations, there is not a graphical editor for creating and modifying these files at this time. |
DIRSIG4 XML Simulation Files
The DIRSIG4 era XML simulation file contained a specific list of input files to describe the single scene to be used, the atmospheric conditions for that scene and the sensor collection description.
<simulation>
<scene externalfile="./demo.scene" />
<atmosphericconditions externalfile="./demo.atm" />
<platform externalfile="./demo.platform" />
<platformmotion externalfile="./demo.ppd" />
<tasklist externalfile="./demo.tasks" />
</simulation>
To provide backward compatibility with DIRSIG4 era simulations, the DIRSIG5 main executable has an XML input simulation file parser that configures the appropriate plugins that provide DIRSIG4 capabilities. Under the hood, the BasicAtmosphere plugin handles DIRSIG4 atmospheric modeling responsibilities and the BasicPlatform plugin handles the data collection responsibilities.
DIRSIG5 JSON Simulation Files
In addition to using the plugins distributed with DIRSIG5, end users will eventually be able to create their own plugins to use with their simulations. In order to directly configure which plugins are to be used and what inputs should be supplied to them, a new simulation input file was needed. The new simulation input file is a JSON formatted document that contains a list of scenes and a list of plugins to be utilized in the simulation. The example below is the JSON equivalent of the XML simulation file example from the last section:
[{
"scene_list" : [
{ "inputs" : "./demo.scene" }
],
"plugin_list" : [
{
"name" : "BasicAtmosphere",
"inputs" : {
"atmosphere_filename" : "./demo.atm"
}
},
{
"name" : "BasicPlatform",
"inputs" : {
"platform_filename" : "./demo.platform",
"motion_filename" : "./demo.ppd",
"tasks_filename" : "./demo.tasks"
}
}
]
}]
The top-level components of the file are the "scene list" and "plugin list". These will be described further in the following sections.
Scene List
One of the features of DIRSIG5 is that it can utilize multiple scenes
in a simulation. This gives scene builders more flexibility when
constructing large or complex scenes and it gives the model the ability
to progressively load scenes into a simulation. The scene_list
lets
the user define the list of scenes that will be used in a simulation.
Each scene entry in the array has the following variables:
inputs
(required)-
The filename for the input scene file. This can be either the
.scene
file or the.scene.hdf
file.
name
(optional)-
A user-defined name associated with the scene. This can make simulations that use many scenes more manageable. Specifically, it makes it possible to request material abundance truth for a specific material in a specific scene.
offset
(optional)-
The 2D offset (the Z is ignored) in the Scene ENU coordinate system for a scene relative to the first scene defined (see below).
Combining Scenes
In order to avoid the headaches of manipulating massive scene geometry
models, a common approach to large-area scene construction has been to
divide the scene into multiple "parts" or "tiles". This is the approach
that has been used at RIT for scenes including MegaScene1. In the case
of MegaScene1, each tile of the five tiles were constructed using a
shared Scene ENU coordinate system, which means each scene file has
the same geographic location for the Scene ENU origin. In DIRSIG4,
these scene tiles had to be combined by having either a GLIST or scene
file combine the geometry for the desired scenes (e.g. tile_1.scene
,
tile_1_2.scene
, tile_1_2_3.scene
, etc.). In DIRSIG5, these scenes
can be combined at simulation time by providing the list of scene files
that describe each tile.
"scene_list" : [
{ "name" : "tile1", "inputs" : "MegaScene1/tile1.scene" },
{ "name" : "tile2", "inputs" : "MegaScene1/tile2.scene" },
{ "name" : "tile3", "inputs" : "MegaScene1/tile3.scene" },
{ "name" : "tile4", "inputs" : "MegaScene1/tile4.scene" },
{ "name" : "tile5", "inputs" : "MegaScene1/tile5.scene" }
The user can also utilize this approach to combine scenes that do not share a common Scene ENU coordinate system. In the example below, we combine the Urban scene (downtown Rochester, NY) with MegaScene1 (a suburb just north of Rochester, NY) into a single scene list.
"scene_list" : [
{ "inputs" : "Urban/summer_day.scene" },
{ "inputs" : "MegaScene1/tile1.scene" },
{ "inputs" : "MegaScene1/tile2.scene" },
{ "inputs" : "MegaScene1/tile3.scene" },
{ "inputs" : "MegaScene1/tile4.scene" },
{ "inputs" : "MegaScene1/tile5.scene" }
To be clear about what is happening here, the Urban scene uses a geographic location to tie its local Scene ENU coordinate system to the Earth in one location and the MegaScene1 scene has it’s own geographic location that ties its local Scene ENU coordinate system to another location on the Earth. These two scenes are near each other in the real world, but these scenes don’t actually meet at the edges. Hence, there is a gap between them. If a wide-area image of the Rochester region was generated, both scenes would be visible at there respective locations in the simulation output.
Because of the precision required in the geolocation, it can be difficult to align separate scenes to cleanly meet at the edges without gaps. Therefore, when constructing a scene using a tiling method, it is strongly encouraged to use a shared coordinate system and a single geolocation for each tile. However an alternative method is also available, which is discussed next.
Relative Offsets
When relying on the geographic location of the location scene origin in
individual scenes to perform relative layouts, the user can quickly get
frustrated with precision and projection issues associated with the
manipulating geographic locations in order to align scene edges. To
simplify that process, individual scenes can be constructed in a single
local Scene ENU coordinate system and then offset relative to each
other when provided to the simulation. The ModularCity1 demo uses
this approach to construct a simple city by quilting together an 11 x 11
set of modular city building blocks that are each 100 x 100 meters. The
offset
for each scene specifies where that scene is relative to the
first scene in the list.
"scene_list" : [
{ "inputs" : "./scenes/city4.scene" },
{ "offset" : "-500,-500,0", "inputs" : "./scenes/tile9.scene" },
{ "offset" : "-500,-400,0", "inputs" : "./scenes/park2.scene" },
{ "offset" : "-500,-300,0", "inputs" : "./scenes/tile4.scene" },
{ "offset" : "-500,-200,0", "inputs" : "./scenes/tile11.scene" },
{ "offset" : "-500,-100,0", "inputs" : "./scenes/tile3.scene" },
{ "offset" : "-500,0,0", "inputs" : "./scenes/tile5.scene" },
{ "offset" : "-500,100,0", "inputs" : "./scenes/park3.scene" },
{ "offset" : "-500,200,0", "inputs" : "./scenes/tile2.scene" },
{ "offset" : "-500,300,0", "inputs" : "./scenes/tile3.scene" },
{ "offset" : "-500,400,0", "inputs" : "./scenes/tile8.scene" },
{ "offset" : "-500,500,0", "inputs" : "./scenes/tile6.scene" },
{ "offset" : "-400,-500,0", "inputs" : "./scenes/park1.scene" },
...
{ "offset" : "400,500,0", "inputs" : "./scenes/tile7.scene" },
{ "offset" : "500,-500,0", "inputs" : "./scenes/tile12.scene" },
{ "offset" : "500,-400,0", "inputs" : "./scenes/park2.scene" },
{ "offset" : "500,-300,0", "inputs" : "./scenes/tile8.scene" },
{ "offset" : "500,-200,0", "inputs" : "./scenes/tile8.scene" },
{ "offset" : "500,-100,0", "inputs" : "./scenes/tile8.scene" },
{ "offset" : "500,0,0", "inputs" : "./scenes/tile5.scene" },
{ "offset" : "500,100,0", "inputs" : "./scenes/tile12.scene" },
{ "offset" : "500,200,0", "inputs" : "./scenes/tile5.scene" },
{ "offset" : "500,300,0", "inputs" : "./scenes/tile6.scene" },
{ "offset" : "500,400,0", "inputs" : "./scenes/tile8.scene" },
{ "offset" : "500,500,0", "inputs" : "./scenes/tile1.scene" }
],
Important Notes:
-
The first scene in the
scene_list
is considered the "anchor" scene. Anoffset
assigned to this scene is ignored.-
Therefore, it follows that an
offset
applied to a single scene simulation would be ignored.
-
-
The although the
offset
is assigned an XYZ triplet, only the X and Y values are used. The Z value of the offset is ignored.
Overlaying Scenes
The previous discussion focused on "tiling" approaches, where individual
scenes do not overlap. However, combined scenes can overlap as well.
Consider a scenario where the user creates a city and is modeling different
traffic patterns through the city. Rather than have a single large scene
that combines the baseline city geometry (buildings, roads, etc.) and a
specific state of traffic, the user could separate the baseline city from
the traffic. The city.scene
would contain only the baseline city
geometry (buildings, roads, etc.) and separate "traffic" scenes would
contain only the vehicles and dynamic instance data describing the path
those vehicles follow through the city. Each traffic scene could capture
a different traffic pattern scenario and the city and traffic can be
combined at simulation time (see below):
"scene_list" : [
{ "inputs" : "/city.scene" },
{ "inputs" : "/traffic_weekday.scene" }
],
Alternatively the user can combine that same city scene with a less busy traffic pattern for the weekend.
"scene_list" : [
{ "inputs" : "/city.scene" },
{ "inputs" : "/traffic_weekend.scene" }
],
Since the cars are driving along roads in the city, it would be easiest to use a shared Scene ENU coordinate system when constructing the scenes.
Plugin List
The second major component to a simulation is the list of
plugins that will be used in the simulation.
This list contains a group of JSON objects for each plugin to be
loaded. Each JSON object includes a name
and inputs
array. The
text assigned to the name
is assumed to match the name advertised
by a given plugin (part of it’s implementation). This name is used
to find the requested plugin in the folders configured in the plugin
search path. The inputs
array is passed directly to the plugin,
hence, the contents of that array are unique to each plugin.
Plugins are searched for in specific locations. More information on how this search is performed is described here. |
For the DIRSIG4 backward compatibility plugins, the inputs
array
contains the name of one or more input files used by the plugin to
configure itself. The example below shows that the
BasicAtmosphere plugin is expecting to
find the atmosphere_filename
object to find the name of the DIRSIG4
era atmosphere file to configure itself from. The
BasicPlatform plugin is expecting
to see the names of the DIRSIG4 era platform, motion and tasks files
to configure itself from.
"plugin_list" : [
{
"name" : "BasicAtmosphere",
"inputs" : {
"atmosphere_filename" : "uniform.atm"
}
},
{
"name" : "BasicPlatform",
"inputs" : {
"platform_filename" : "demo.platform",
"motion_filename" : "demo.ppd",
"tasks_filename" : "demo.tasks"
}
}
]
However, there is nothing stopping a developer from creating a plugin that
embeds the input parameters directly into the inputs
array rather than
read them from an external file.
"plugin_list" : [
{
"name" : "MyAtmospherePlugin",
"inputs" : {
"sun_irradiance" : 100.0,
"sky_irradiance" : 20.0,
"sun_zenith" : 83.21,
"sun_azimuth" : 210.8
}
},
...
]
Combination Plugins
Some plugins in DIRSIG5 are "combination plugins", which means that a single plugin implements several plugin types. For example, the ChipMaker plugin is a sensor plugin, an atmosphere plugin and an ephemeris plugin.
Multiple Plugins and Overriding Plugins
The plugin list allows the user to configure multiple plugins to be used in the simulation. Some plugin types allow multiple instances of the plugin to be used in the simulation. However, for a subset of plugin types, DIRSIG5 assumes there will only be one plugin supplied. The table below outlines which plugin types support multiple instances.
Type | Single vs. Multiple |
---|---|
Atmosphere |
Single |
Ephemeris |
Single |
Sensor |
Multiple |
Weather |
Single |
It is possible that a plugin list will include more than one plugin
for a plugin type that only supports a single instance of that plugin.
For example, the user might mistakenly configure one or more weather
plugins. When this occurs, subsequent plugin requests for that plugin
type will override the previous plugin request. For example, if
the user requests a weather plugin named WeatherA
and then WeatherB
,
the WeatherB
plugin is the one that will be used in the simulation.
The simulation startup will communicate this override in the configuration:
Setting up the 'WeatherA' plugin Setting up the 'WeatherB' plugin Replacing 'WeatherA' with 'WeatherB' as the weather plugin
It is not uncommon to deliberately configure plugin override
scenarios. For example, you might request the BasicAtmosphere
plugin and provide it a DIRSIG4-era .atm file, which will
automatically configure instances of the BasicAtmosphere
plugin, the SpiceEphemeris plugin and the ThermWeather
plugin. However, you might want to deliberately override the
older ThermWeather plugin with the newer NsrdbWeather
plugin. In that case, make sure you request the NsrdbWeather
plugin after the BasicAtmosphere plugin.
|
Medium List
In addition to providing a generic way to load plugins into the
simulation, the JSIM format has specific definitions to facilitate
setting up simulation components that might be composed of multiple
plugins and related options. The primary example of this is a
medium_list
that defines individual medium components that have
plugins for optical properties, the boundary of the medium, and
possibly medium-based geometry. A medium_list
exists at the same
level in the JSIM as the scene_list
and plugin_list
and has
three base components:
[{
"scene_list" : [ ... ],
"medium_list" : [
{
"type" : <type of medium>,
"plugin_list" : [
<plugin entries>
],
"options" : {
<options specific to the type of medium>
}
}
],
"plugin_list" : [ .. ]
}]
The exact definitions of these sections are dependent on the type
of medium being described and are documented elsewhere (for example,
the water plugins and
plume plugins). The medium_list
is completely
optional and will not be found in most JSIM files.
Using a JSIM file
The DIRSIG5 main executable checks to see if the supplied input file is a DIRSIG4 SIM file or a DIRSIG5 JSIM file and then parse it appropriately. Hence, using a JSIM file is as easy as:
$ dirsig5 demo.jsim