6. ROS Synchronization (Linux Only)

This tutorial covers the use of the ROS synchronization tools constituting the communication to and from the purple layers (perception and control) in the Cortex pipeline.

Isaac Cortex pipeline.

See Isaac Cortex: Overview for more detail on the Cortex pipeline. The setup of the Cortex world and behaviors is largely the same with and without ROS communication, we just need to create some ROS-specific objects that handle communication in the background. We’ll see in the final section that all Franka behaviors can be run with the example_cortex_sync_main.py example similar to franka_examples_main.py.

In all command line examples, we use the abbreviation isaac_python for the Isaac Sim python script (<isaac_sim_root>/python.sh on Linux and <isaac_sim_root>\python.bat on Windows). The command lines are written relative to the working directory standalone_examples/api/omni.isaac.cortex since many scripts reference the behaviors subdirectory.

Each example will launch Isaac Sim without playing the simulation. Press play to run the simulation and behavior.

6.1. Preliminaries: Build the Cortex control ROS package

For this tutorial, we need to build the cortex_control ROS package. We’ll give instructions here for building the complete noetic_ws that comes with Isaac Sim ROS repository, but for a minimal workspace, you can create your own (or use an existing workspace) and simply copy or symlink cortex_control into it.

Building cortex_control:

source /opt/ros/noetic/setup.bash

cd <noetic_ws>

source devel/setup.bash
roscd cortex_control  # This should get you into the package directory.

6.2. Block stacking with ROS synchronization

Start a roscore.

Run the following demo

isaac_python example_cortex_sync_main.py

The example will launch a Franka scene with colored blocks off to the side and the block stacking behavior. There will be two Frankas, one in front and one in back. The Franka scene in front represents the mind of the robot, its belief. The one in back is a simulated version of reality. See Isaac Cortex: Overview for details on the separation between belief and simulated reality.

We will refer to the robot in front as the belief robot and the robot in back as the sim robot.

Press play.

You will see the belief robot attempt to pick up a block, but fail. The belief robot moves but the sim robot does not because the two aren’t connected by a controller. The blocks are being synchronized since perception is simulated with ground truth transforms. Try moving the sim blocks (in the back). The blocks in the front move as well. Note that the block updates are throttled in this example so you’ll see the blocks synchronize only at 4hz (every 25 ms). When the belief attempts to pick a block, it’ll think it’s successful for around a quarter of a second, then realize the “real” block in the sim environment hasn’t moved and synchronize the belief back to the sim “reality”.

Start the sim controller:

rosrun cortex_control sim_controller

You’ll see the belief robot synchronize with the sim robot’s configuration and start controlling it. The sim robot will move along with the belief robot and successfully accomplish the block stacking task in the sim world. You can interact with the system similar to the non-synchronized block stacking example by interacting with the sim world in back.

With the simulated controller running it’s safe to pause or stop (reset) the simulation. The controller will automatically re-connect when necessary.

6.3. Details

Take a look at the code for the sync example in standalone_examples/api/omni.isaac.cortex/example_cortex_sync_main.py. It’s very similar to the code in the non-synchronized example standalone_examples/api/omni.isaac.cortex/franka_examples_main.py. The main differences are three-fold:

  1. We create a second robot and set of blocks (the sim environment). You can think of the single environment in the non-synchronized example as a belief environment.

  2. The belief objects are wrapped as CortexObject objects. This class has a member that will store the measured object poses and methods that facilitate synchronizing the object’s pose with the measured pose.

  3. We create a collection of four communication objects (described below).

Other than that, the setup of the CortexWorld and the behavior is the same.

The communication objects are:

cortex_sim = CortexSimRobotRos(sim_robot)
cortex_sim_objects_ros = CortexSimObjectsRos(sim_objects)
cortex_control = CortexControlRos(robot)
cortex_objects_ros = CortexObjectsRos(cortex_objects, auto_sync_objects=args.auto_sync_objects)

These objects do the following:

  • CortexSimRobotRos creates a ROS interface to the simulated robot. It handles listening to low-level commanded joint values simulating low-level control. These low-level commands will be sent by the simulated controller. It also handles receiving and processing gripper commands.

  • CortexSimObjectsRos handles publishing the “measured” object poses simulating perception.

  • CortexControlRos handles sending commands to the sim robot. Note that gripper commands go directly to the sim robot ROS node, but arm commands are sent to the controller which handles the synchronization handshake and then then relays interpolated versions of the commands to the sim robot’s ROS interface.

  • CortexObjectsRos handles listening to the measured object poses and recording them in the CortexObject objects.

The CortexControlRos and CortexObjectsRos objects are agnostic to whether they’re talking to the simulated ROS interface (using CortexSimRobotRos and CortexSimObjectsRos) or to the same ROS interface on the real robot.

6.4. Running the other behaviors using ROS synchronization

By default, the CortexObject objects record the measured poses but don’t automatically synchronize the object poses to those measured poses. Instead, the user or behavior should decide how and when that synchronization should take place based on the needs of the behavior and the characteristics of the perception module.

The block stacking behavior gives an example of handling that synchronization in the behavior. Specifically, it’s handled in the BuildTowerContex.monitor_perception monitor.

Alternatively, the CortexObjectsRos object can be configured to automatically synchronize the belief blocks every cycle to their measured poses on construction:

cortex_objects_ros = CortexObjectsRos(cortex_objects, auto_sync_objects=True)

In the example standalone_examples/api/omni.isaac.cortex/example_cortex_sync_main.py, that setting is activated using the flag --auto_sync_objects.

Any of the behaviors listed in standalone_examples/api/omni.isaac.cortex/behaviors/franka can be loaded with this example. With all examples except the block stacking behavior the flag --auto_sync_objects should be used.

The full commandline is

isaac_python example_cortex_sync_main.py --behavior=<path_to_behavior> --auto_sync_objects

with <path_to_behavior> set to any of the following


Note that the block stacking behavior behaviors/franka/block_stacking_behavior.py (which the flat --behavior defaults to) is left off the list because it shouldn’t use the --auto_sync_objects flag.

6.4.1. Interacting with the sim env

When the --auto_sync_objects flag is used, the belief objects (in front) will be synchronized with the sim objects (in back) every step, therefore you won’t be able to manually move them. Instead, move the sim objects in back and you’ll see the belief objects in front move.

Try zooming in to the sim scene by clicking on the base of the back robot and pressing f to focus the camera there. Adjust the scene by holding down Alt, dragging the mouse and using the scrollwheel to zoom as needed. This perspective, showing only the sim robot, mimics what you’d experience in the real world. The belief robot is making decisions, but you’re watching the sim robot and interacting with the sim environment.

6.5. Controlling physical robots

Controlling a physical robot requires that you have a controller implementing the connection and communication protocol required for syncing with Cortex. The cortex_control ROS package provides a reference implementation of the protocol in a command stream interpolation tool used inside the sim_controller described above. This tool can be used inside real controllers to handle synchronizing Cortex with the physical robot and smoothly controlling the robot.

The header and implementation files


contain a class CommandStreamInterpolator that encapsulates an interpolator that can connect with Cortex via ROS following the required connection and synchronization protocol. The file


implements the sim_controller and demonstrates how to use those tools. The ROS package cortex_control_franka provides a reference implementation of a real-world Franka controller using the CommandStreamInterpolator class.


To run Cortex to control a physical Franka you need a physical Franka robot with Franka Control Interface (FCI) and franka_ros. See:

along with a real-time control workstation setup as described here:

Without that setup, the cortex_control_franka commands will error.

To launch the system for controlling a physical robot you’ll launch a single Franka environment as you usually would, but create ROS communication objects for the belief robot as shown in the example_cortex_sync_belief_main.py example:

cortex_control = CortexControlRos(robot)
cortex_objects_ros = CortexObjectsRos(cortex_objects, auto_sync_objects=args.auto_sync_objects)

Note that you don’t need to create the simulated robot and objects or the CortexSimRobotRos and CortexSimObjectsRos objects. The physical robot, an external perception module, and the physical robot’s controller (described below) play those roles.

6.5.1. Starting the Franka controllers

We describe here how to start up the Franka controllers which play the role of the real-world variant of the sim_controller.

Note that, in the instructions below, at the point were we launch the joint position controller in Terminal 3 below, you should see the Cortex belief robot synchronize with the physical robot. It will engage the physical robot at that point, so you may see some slight movement, but it shouldn’t be much since the belief has synchronized with the physical robot’s configuration. However, make sure you have the e-stop ready in case anything goes wrong.


Make sure you have the e-stop in hand when running the following commands in case anything goes wrong. In particular, the command in Terminal 3 (starting the position controller) will engage the physical robot.

# Terminal 1: Start the Franka controller manager
roslaunch cortex_control_franka franka_control_cortex.launch

# Terminal 2: Set high torque thresholds for Franka
rosrun cortex_control_franka set_high_collision_thresholds

# Terminal 3: Startup the position controller -- launching this controller syncs the belief with the physical robot.
roslaunch cortex_control_franka joint_position_controller.launch

# Terminal 4: Start the gripper commander listener
rosrun cortex_control_franka franka_gripper_command_relay.py


You will have to source the appropriate ROS workspace to run the above commands.

You can run the behaviors you ran in simulation in this way on the physical robot. The simplest version for testing is to load no behavior and simply manually control the end-effector using the /World/motion_commander_target prim.

6.5.2. Simple simulated demo

We don’t give a full demo of launching the physical robot controllers, but we mimic it using the simulated setup used above. Here, rather than launching both the belief and sim robots / environments in the same instance of Isaac Sim, we use two separate instances of Isaac Sim, one for the belief world and one for the sim world. This is analogous to the setup you’d use for a physical robot. The main difference in a physical setup is, rather than launch a separate instance of Isaac Sim to represent the sim world, we’d use the physical world and robot.

Launch an instance of Isaac Sim to represent the simulated reality (sim world):

isaac_python example_cortex_sync_sim_main.py

There’s no behavior associated with the sim world since it represents the real world. Now launch the belief world:

isaac_python example_cortex_sync_belief_main.py

By default, it launches no behavior (we’ll manually control the end-effector). Launch the sim_controller as before using

rosrun cortex_control sim_controller

With the sim_controller running, the two robots should be synchronized and connected, although you might not see either move since they both launch into the same configuration. Try manually controlling the belief robot using the /World/motion_commander_target prim.

Behaviors can be added to the belief robot as usual.

This demo can be run with the separate instances of Isaac Sim running on separate machines. Simply setup ROS to communicate between the machines and follow the instructions above. Run the sim_controller from the machine running the the sim world.