Real-time simulation

Hello. My question is: How do the physics calculation step and the rendering step relate to real time? Examples such as from standalone_example/api/omni.isaac.core, contain the following code:

simulation_context.set_simulation_dt(physics_dt=1.0 / 60.0, rendering_dt=1.0 / 600.0)
for step in range(10):

If we define callback functions:

simulation_context.add_physics_callback("physics_callback", step_callback)
simulation_context.add_render_callback("render_callback", render_callback)

We will see that for one call to the step_callback function, there are 10 calls to the render_callback function. If in the render_callback function we will put timestamps:

import time
def render_callback(event):
    print("update app with step: ", time.time())

we will see that the actual frequency of calling the render_callback function does not correspond to 60 Hz.
How do I organize the main loop in which simulation_context.step is called to provide a time close to real time? I need to be able to, for example, publish ROS messages with a certain frequency, etc.

1 Like

When running a standalone example with SimulationApp, Isaac sim will run as fast as possible.
Your options are:

  • publish a ROS Clock message and use that as your time so things stay in sync"
  • use time.sleep between each step call to get it to run at a slower rate. This works if your simulation is running faster than realtime

one example of the latter: (The SteadyRate class is copied from omni.isaac.cortex

class SteadyRate:
    """ Maintains the steady cycle rate provided on initialization by adaptively sleeping an amount
    of time to make up the remaining cycle time after work is done.


    rate = SteadyRate(rate_hz=30.)
    while True:  # Do any work.
      rate.sleep()  # Sleep for the remaining cycle time.


    def __init__(self, rate_hz):
        self.rate_hz = rate_hz
        self.dt = 1.0 / rate_hz
        self.last_sleep_end = time.time()

    def sleep(self):
        work_elapse = time.time() - self.last_sleep_end
        sleep_time = self.dt - work_elapse
        if sleep_time > 0.0:
        self.last_sleep_end = time.time()

rate = SteadyRate(60)

simulation_context.set_simulation_dt(physics_dt=1.0 / 60.0, rendering_dt=1.0 / 600.0)
for step in range(10):

Hello, @Hammad_M. Thank you so much for your reply. I have a few comments:

  1. When using the ROS Clock block in the model, the time in the other nodes will be synchronized with the time of the model. And it works fine if I don’t have to enter any data from physical devices. For example, I has implemented a car model in Isaac Sim, which works 3 times faster than real time, and I want to control it using the game steering wheel. It means that I will have to drive three times more actively.

  2. I can write the main application loop myself, as you suggest to do, however, I would prefer to run the model from the main Isaac Sim window and have a parameter “Real-time simulation enabled” as for example it is done in CoppeliaSim, for real-time emulation.

Most of our examples use the IsaacReadSimulationTime node for publishing data.
There is also a IsaacReadSystemTime OmniGraph node that can be used as a drop in replacement