Multiple env replicator generation with capturing, half of the envs aren't randomized

Omniverse Isaac Sim
1

offline_example.py (4.0 KB)

Hi,

I’m trying to generate a dataset of objects. I want to be able to control their locations in code, per simulation step. Half of them, (the upper half in terms of env numbers) don’t get randomized with this code. They just perform physics steps, and roll.

Supplementally, how do I actually control each objects position? I want to, for example, put a set of objects in a line or circle.

How can I check for collisions with disabled physics?

How can I register a pose annotator?

How can I unload objects? There’s very little in the way of this in the documentation. I want different sets probably at each step.

I’m also not confident in my approach. Would it be better to forgo the multiple environments, and just re-build one over and over? The cameras can see eachothers environments.

3

Hi there,

if you do not need to run cloned simulations you might be easier off having one environment to collect the data, especially if you do not want the cameras to see the other environments. You can take a look at the replicator extension for this: Getting started with Omniverse Replicator — Omniverse Extensions documentation

Let me know if you require any further help/directions.

Best,
Andrei