Publishing Camera’s Data

Learning Objectives

In this tutorial, we demonstrate how to programmatically set up publishers for Isaac Sim Cameras at an approximate frequency.

Getting Started

Prerequisite

Setup a camera in a scene

To begin this tutorial, we first set up an environment with a omni.isaac.sensor Camera object. Running the below will result in a simple warehouse environment loaded with a camera in the scene.

Run roscore in a separate window, and then run the following script:

 1import carb
 2from isaacsim import SimulationApp
 3import sys
 4
 5BACKGROUND_STAGE_PATH = "/background"
 6BACKGROUND_USD_PATH = "/Isaac/Environments/Simple_Warehouse/warehouse_with_forklifts.usd"
 7
 8CONFIG = {"renderer": "RayTracedLighting", "headless": False}
 9
10# Example ROS bridge sample demonstrating the manual loading of stages and manual publishing of images
11simulation_app = SimulationApp(CONFIG)
12import omni
13import numpy as np
14import omni.graph.core as og
15import omni.replicator.core as rep
16import omni.syntheticdata._syntheticdata as sd
17
18from omni.isaac.core import SimulationContext
19from omni.isaac.core.utils import stage, extensions, nucleus
20from omni.isaac.sensor import Camera
21import omni.isaac.core.utils.numpy.rotations as rot_utils
22from omni.isaac.core.utils.prims import is_prim_path_valid
23from omni.isaac.core_nodes.scripts.utils import set_target_prims
24
25# Enable ROS bridge extension
26extensions.enable_extension("omni.isaac.ros_bridge")
27
28simulation_app.update()
29
30# check if rosmaster node is running
31# this is to prevent this sample from waiting indefinetly if roscore is not running
32# can be removed in regular usage
33import rosgraph
34
35if not rosgraph.is_master_online():
36    carb.log_error("Please run roscore before executing this script")
37    simulation_app.close()
38    exit()
39
40simulation_context = SimulationContext(stage_units_in_meters=1.0)
41
42# Locate Isaac Sim assets folder to load environment and robot stages
43assets_root_path = nucleus.get_assets_root_path()
44if assets_root_path is None:
45    carb.log_error("Could not find Isaac Sim assets folder")
46    simulation_app.close()
47    sys.exit()
48
49# Loading the environment
50stage.add_reference_to_stage(assets_root_path + BACKGROUND_USD_PATH, BACKGROUND_STAGE_PATH)
51
52###### Camera helper functions for setting up publishers. ########
53
54# Paste functions from the tutorial here
55# def publish_camera_tf(camera: Camera): ...
56# def publish_camera_info(camera: Camera, freq): ...
57# def publish_pointcloud_from_depth(camera: Camera, freq): ...
58# def publish_depth(camera: Camera, freq): ...
59# def publish_rgb(camera: Camera, freq): ...
60
61###################################################################
62
63# Create a Camera prim. Note that the Camera class takes the position and orientation in the world axes convention.
64camera = Camera(
65    prim_path="/World/floating_camera",
66    position=np.array([-3.11, -1.87, 1.0]),
67    frequency=20,
68    resolution=(256, 256),
69    orientation=rot_utils.euler_angles_to_quats(np.array([0, 0, 0]), degrees=True),
70)
71camera.initialize()
72
73simulation_app.update()
74camera.initialize()
75
76############### Calling Camera publishing functions ###############
77
78# Call the publishers.
79# Make sure you pasted in the helper functions above, and uncomment out the following lines before running.
80
81approx_freq = 30
82#publish_camera_tf(camera)
83#publish_camera_info(camera, approx_freq)
84#publish_rgb(camera, approx_freq)
85#publish_depth(camera, approx_freq)
86#publish_pointcloud_from_depth(camera, approx_freq)
87
88####################################################################
89
90# Initialize physics
91simulation_context.initialize_physics()
92simulation_context.play()
93
94while simulation_app.is_running():
95    simulation_context.step(render=True)
96
97simulation_context.stop()
98simulation_app.close()

Publish camera intrinsics to CameraInfo topic

The following snippet will publish camera intrinsics associated with an omni.isaac.sensor Camera to a sensor_msgs/CameraInfo topic.

 1def publish_camera_info(camera: Camera, freq):
 2    # The following code will link the camera's render product and publish the data to the specified topic name.
 3    render_product = camera._render_product_path
 4    step_size = int(60/freq)
 5    topic_name = camera.name+"_camera_info"
 6    queue_size = 1
 7    node_namespace = ""
 8    frame_id = camera.prim_path.split("/")[-1] # This matches what the TF tree is publishing.
 9
10    stereo_offset = [0.0, 0.0]
11
12    writer = rep.writers.get("ROS1PublishCameraInfo")
13    writer.initialize(
14        frameId=frame_id,
15        nodeNamespace=node_namespace,
16        queueSize=queue_size,
17        topicName=topic_name,
18        stereoOffset=stereo_offset,
19    )
20    writer.attach([render_product])
21
22    gate_path = omni.syntheticdata.SyntheticData._get_node_path(
23        "PostProcessDispatch" + "IsaacSimulationGate", render_product
24    )
25
26    # Set step input of the Isaac Simulation Gate nodes upstream of ROS publishers to control their execution rate
27    og.Controller.attribute(gate_path + ".inputs:step").set(step_size)
28    return

Publish Depth

 1def publish_pointcloud_from_depth(camera: Camera, freq):
 2    # The following code will link the camera's render product and publish the data to the specified topic name.
 3    render_product = camera._render_product_path
 4    step_size = int(60/freq)
 5    topic_name = camera.name+"_pointcloud" # Set topic name to the camera's name
 6    queue_size = 1
 7    node_namespace = ""
 8    frame_id = camera.prim_path.split("/")[-1] # This matches what the TF tree is publishing.
 9
10    # Note, this pointcloud publisher will simply convert the Depth image to a pointcloud using the Camera intrinsics.
11    # This pointcloud generation method does not support semantic labelled objects.
12    rv = omni.syntheticdata.SyntheticData.convert_sensor_type_to_rendervar(
13        sd.SensorType.DistanceToImagePlane.name
14    )
15
16    writer = rep.writers.get(rv + "ROS1PublishPointCloud")
17    writer.initialize(
18        frameId=frame_id,
19        nodeNamespace=node_namespace,
20        queueSize=queue_size,
21        topicName=topic_name
22    )
23    writer.attach([render_product])
24
25    # Set step input of the Isaac Simulation Gate nodes upstream of ROS publishers to control their execution rate
26    gate_path = omni.syntheticdata.SyntheticData._get_node_path(
27        rv + "IsaacSimulationGate", render_product
28    )
29    og.Controller.attribute(gate_path + ".inputs:step").set(step_size)
30
31    return

Publish RGB images

 1def publish_rgb(camera: Camera, freq):
 2    # The following code will link the camera's render product and publish the data to the specified topic name.
 3    render_product = camera._render_product_path
 4    step_size = int(60/freq)
 5    topic_name = camera.name+"_rgb"
 6    queue_size = 1
 7    node_namespace = ""
 8    frame_id = camera.prim_path.split("/")[-1] # This matches what the TF tree is publishing.
 9
10    rv = omni.syntheticdata.SyntheticData.convert_sensor_type_to_rendervar(sd.SensorType.Rgb.name)
11    writer = rep.writers.get(rv + "ROS1PublishImage")
12    writer.initialize(
13        frameId=frame_id,
14        nodeNamespace=node_namespace,
15        queueSize=queue_size,
16        topicName=topic_name
17    )
18    writer.attach([render_product])
19
20    # Set step input of the Isaac Simulation Gate nodes upstream of ROS publishers to control their execution rate
21    gate_path = omni.syntheticdata.SyntheticData._get_node_path(
22        rv + "IsaacSimulationGate", render_product
23    )
24    og.Controller.attribute(gate_path + ".inputs:step").set(step_size)
25
26    return

Publish depth images

 1def publish_depth(camera: Camera, freq):
 2    # The following code will link the camera's render product and publish the data to the specified topic name.
 3    render_product = camera._render_product_path
 4    step_size = int(60/freq)
 5    topic_name = camera.name+"_depth"
 6    queue_size = 1
 7    node_namespace = ""
 8    frame_id = camera.prim_path.split("/")[-1] # This matches what the TF tree is publishing.
 9
10    rv = omni.syntheticdata.SyntheticData.convert_sensor_type_to_rendervar(
11                            sd.SensorType.DistanceToImagePlane.name
12                        )
13    writer = rep.writers.get(rv + "ROS1PublishImage")
14    writer.initialize(
15        frameId=frame_id,
16        nodeNamespace=node_namespace,
17        queueSize=queue_size,
18        topicName=topic_name
19    )
20    writer.attach([render_product])
21
22    # Set step input of the Isaac Simulation Gate nodes upstream of ROS publishers to control their execution rate
23    gate_path = omni.syntheticdata.SyntheticData._get_node_path(
24        rv + "IsaacSimulationGate", render_product
25    )
26    og.Controller.attribute(gate_path + ".inputs:step").set(step_size)
27
28    return

Publish a TF Tree for the camera pose

The pointcloud, published using the above function, will publish the pointcloud in the ROS camera axes convention (-Y up, +Z forward). To make visualizing this pointcloud easy in ROS via RViz, the following snippet will publish a TF Tree to the /tf, containing two frames.

The two frames are:

  1. {camera_frame_id}: This is the camera’s pose in the ROS camera convention (-Y up, +Z forward). Pointclouds are published in this frame.

    ../_images/camera_frames_v2.005.png

  2. {camera_frame_id}_world: This is the camera’s pose in the World axes convention (+Z up, +X forward). This will reflect the true pose of the camera.

    ../_images/camera_frames_v2.004.png

The TF Tree looks like this:

../_images/transformation.png

  • world -> {camera_frame_id} is a dynamic transform from origin to the camera in the ROS camera convention, following any movement of the camera.

  • {camera_frame_id} -> {camera_frame_id}_world is a static transform consisting of only a rotation and zero translation. This static transform can be represented by the quaternion [0.5, -0.5, 0.5, 0.5] in [w, x, y, z] convention.

Since the pointcloud is published in {camera_frame_id}, it is encouraged to set the frame_id of the pointcloud topic to {camera_frame_id}. The resulting visualization of the pointclouds can be viewed in the world frame in RViz.

 1def publish_camera_tf(camera: Camera):
 2    camera_prim = camera.prim_path
 3
 4    if not is_prim_path_valid(camera_prim):
 5        raise ValueError(f"Camera path '{camera_prim}' is invalid.")
 6
 7    try:
 8        # Generate the camera_frame_id. OmniActionGraph will use the last part of
 9        # the full camera prim path as the frame name, so we will extract it here
10        # and use it for the pointcloud frame_id.
11        camera_frame_id=camera_prim.split("/")[-1]
12
13        # Generate an action graph associated with camera TF publishing.
14        ros_camera_graph_path = "/CameraTFActionGraph"
15
16        # If a camera graph is not found, create a new one.
17        if not is_prim_path_valid(ros_camera_graph_path):
18            (ros_camera_graph, _, _, _) = og.Controller.edit(
19                {
20                    "graph_path": ros_camera_graph_path,
21                    "evaluator_name": "execution",
22                    "pipeline_stage": og.GraphPipelineStage.GRAPH_PIPELINE_STAGE_SIMULATION,
23                },
24                {
25                    og.Controller.Keys.CREATE_NODES: [
26                        ("OnTick", "omni.graph.action.OnTick"),
27                        ("IsaacClock", "omni.isaac.core_nodes.IsaacReadSimulationTime"),
28                        ("RosPublisher", "omni.isaac.ros_bridge.ROS1PublishClock"),
29                    ],
30                    og.Controller.Keys.CONNECT: [
31                        ("OnTick.outputs:tick", "RosPublisher.inputs:execIn"),
32                        ("IsaacClock.outputs:simulationTime", "RosPublisher.inputs:timeStamp"),
33                    ]
34                }
35            )
36
37        # Generate 2 nodes associated with each camera: TF from world to ROS camera convention, and world frame.
38        og.Controller.edit(
39            ros_camera_graph_path,
40            {
41                og.Controller.Keys.CREATE_NODES: [
42                    ("PublishTF_"+camera_frame_id, "omni.isaac.ros_bridge.ROS1PublishTransformTree"),
43                    ("PublishRawTF_"+camera_frame_id+"_world", "omni.isaac.ros_bridge.ROS1PublishRawTransformTree"),
44                ],
45                og.Controller.Keys.SET_VALUES: [
46                    ("PublishTF_"+camera_frame_id+".inputs:topicName", "/tf"),
47                    # Note if topic_name is changed to something else besides "/tf",
48                    # it will not be captured by the ROS tf broadcaster.
49                    ("PublishRawTF_"+camera_frame_id+"_world.inputs:topicName", "/tf"),
50                    ("PublishRawTF_"+camera_frame_id+"_world.inputs:parentFrameId", camera_frame_id),
51                    ("PublishRawTF_"+camera_frame_id+"_world.inputs:childFrameId", camera_frame_id+"_world"),
52                    # Static transform from ROS camera convention to world (+Z up, +X forward) convention:
53                    ("PublishRawTF_"+camera_frame_id+"_world.inputs:rotation", [0.5, -0.5, 0.5, 0.5]),
54                ],
55                og.Controller.Keys.CONNECT: [
56                    (ros_camera_graph_path+"/OnTick.outputs:tick",
57                        "PublishTF_"+camera_frame_id+".inputs:execIn"),
58                    (ros_camera_graph_path+"/OnTick.outputs:tick",
59                        "PublishRawTF_"+camera_frame_id+"_world.inputs:execIn"),
60                    (ros_camera_graph_path+"/IsaacClock.outputs:simulationTime",
61                        "PublishTF_"+camera_frame_id+".inputs:timeStamp"),
62                    (ros_camera_graph_path+"/IsaacClock.outputs:simulationTime",
63                        "PublishRawTF_"+camera_frame_id+"_world.inputs:timeStamp"),
64                ],
65            },
66        )
67    except Exception as e:
68        print(e)
69
70    # Add target prims for the USD pose. All other frames are static.
71    set_target_prims(
72        primPath=ros_camera_graph_path+"/PublishTF_"+camera_frame_id,
73        inputName="inputs:targetPrims",
74        targetPrimPaths=[camera_prim],
75    )
76    return

Running the Example

Start a roscore in a separate terminal, save the above script and run using python.sh in the Isaac Sim folder. In our example, {camera_frame_id} is the prim name of the camera, which is floating_camera.

You will see a floating camera with prim path /World/floating_camera in the scene, and the camera should see a forklift:

../_images/isaac_tutorial_ros_camera_publishing_simview.png

If you open a terminal and type rostopic list, you should expect to see the following:

rostopic list
/floating_camera_camera_info
/floating_camera_depth
/floating_camera_pointcloud
/floating_camera_rgb
/clock
/rosout
/rosout_agg
/tf

If we inspect one of the topics, we will see that the topic is publishing approximately at the frequency that we set:

rostopic hz /camera_camera_info
subscribed to [/camera_camera_info]
average rate: 31.298
    min: 0.024s max: 0.039s std dev: 0.00413s window: 31
average rate: 32.119
    min: 0.024s max: 0.041s std dev: 0.00434s window: 64
average rate: 32.817
    min: 0.024s max: 0.041s std dev: 0.00407s window: 98
average rate: 32.551
    min: 0.024s max: 0.041s std dev: 0.00408s window: 130
average rate: 32.509
    min: 0.024s max: 0.041s std dev: 0.00404s window: 162
average rate: 32.589
    min: 0.024s max: 0.041s std dev: 0.00406s window: 195
average rate: 32.423
    min: 0.024s max: 0.041s std dev: 0.00405s window: 227

The frames published by TF will look like the following:

../_images/frames.png

Now, we can visualize the pointcloud and depth images via RViz. Open RViz, and set the “Fixed Frame” field to world.

../_images/rviz.png

Then, enable viewing /floating_camera_depth, /floating_camera_rgb, /floating_camera_pointcloud, and /tf topics.

The depth image /floating_camera_depth and RGB image /floating_camera_rgb should look like this respectively:

../_images/isaac_tutorial_ros_camera_publishing_rgbd.png

The pointcloud will look like so. Note that the camera frames published by the TF publisher shows the two frames. The image on the left shows the {camera_frame_id}_world frame, and the image on the right shows the {camera_frame_id} frame.

../_images/isaac_tutorial_ros_camera_publishing_pc_frontview.png

From the side view:

../_images/isaac_tutorial_ros_camera_publishing_pc_sideview.png

Summary

This tutorial demonstrated how to programmatically set up publishers for Isaac Sim Cameras at an approximate frequency.

Next Steps

Continue on to the next tutorial in our ROS Tutorials series, Ackermann Controller.