General Gazebo capabilities and intent

I am currently looking into autonomous vehicle simulation platforms and I’m trying to get a better sense of where Gazebo fits into this whole space. It seems like a good choice for easy ROS integration, localized prototyping and visualization, not necessarily the best for large-scale system testing of autonomous vehicles. For one, I wasn’t able to find any hi-res urban environment models, except the demo one from 2017 that OSRF created. There aren’t a lot of vehicle models, nor any high-fidelity sensor models. We do use Gazebo to simulate some of our UAVs and I do think it’s a great tool, just not exactly suited for our purposes. What I’m looking for is something along the lines of CARLA or LGSVL. Questions:

  1. Is my assessment of Gazebo generally accurate or is it more capable that I realize?

  2. How does Ignition relate to Gazebo?

Hi @LennyS, thank you for the questions.

You’re talking about CitySim, right? That demo was put together as a proof of concept by 3 engineers in just one week. So it’s definitely not all that Gazebo is capable of, it’s just a taste of it :wink:

Do you have a concrete measure of what you mean by “hi-res”? I assume you’re talking about visual appearance? Gazebo-classic uses Ogre 1.X as the rendering engine, which has been around for a long time. Ignition has a rendering abstraction where any render engine can be plugged in, and we’ve been focusing our efforts on Ogre 2, which supports PBR, for example. So nicer environments can be created in Ignition, like the worlds for the SubT challenge.

Regardless, it’s important to note that just having a fancy rendering engine is not enough - it’s also important to have high quality assets that use all the latest features. Both Gazebo and Ignition support widely available formats like COLLADA, OBJ and STL.

I believe the most elaborate vehicle model we have available for free is the Prius. That model was used in the Prius Challenge Simulator, which was used by teams competing in the physical competition during practice.

Besides that, we have a few rigid vehicles that were used in the demo for surrounding traffic.

Again, could you elaborate on what you mean by high-fidelity? What’s missing on Gazebo? Is it SDF files with specs matching real-world sensors? Better noise models? Different phenomena? Both Gazebo-classic and Ignition support a wide range of sensor capabilities, it would be helpful to understand what you think is missing.

Unlike CARLA and LGSVL, which are specialized simulators for AVs, Gazebo and Ignition are general-purpose simulators. They can be used to simulate AVs, but there are probably many tools and capabilities that you’d get out-of-the-box with the other simulators but not (yet) with Gazebo or Ignition.

I don’t think there’s anything intrinsic about Gazebo or Ignition that prevents the development of such tools though. In fact, there may be groups already building such tools on top of them. I’m not aware of anything fully-fledged and available for free though.

Ignition is a refactor of Gazebo. Perhaps this post answers your question?