Cloud API


Populate realistic scenarios


We drive everything but the ego car


Pre-pay for only the API calls you use.

Volume discounting available.


$0.005 (USD) / call

  • initialize scenarios realistically
  • hundreds of maps and growing


$0.0005 (USD) / call

  • drive 100’s of NPC agents faster than realtime
  • 100 ms. tick
  • <1MB/s network per simulator
  • no scalability cap
  • hundreds of maps and growing
  • left- and right-hand side driving
  • traffic controls aware


1 / Interaction model

High-level documentation about how to integrate INITIALIZE and DRIVE into your simulators is available here, including what, exactly, you need to pass back and forth to our cloud API from your simulation environment.

2 / Python API

Documentation for our Python client library is available here. It is pip installable or installable from source.

3 / C++ API

Documentation for our C++ client library is available here.


Documentation for our low-level REST API is available here.

5 / Examples and integrations

Hands-on examples include Google Colab, example Python scripts, and an example CARLA integration.

Frequently Asked Questions

What about latency?


Latency is only an issue if you can't run your simulation environment in parallel. We measure our packet latency at around 100ms round trip. This will vary depending on where you are in the world relative to our compute, how many agents you drive at once, and other factors. We recommend doubling the number of simulators being run to account for this latency. The improvement in the environment due to the realism, reactivity, and diversity of our NPC behaviors more than makes up for the engineering inconvenience.

What happens if my pre-paid API calls start to run low or run out?


We monitor your API usage and issue warning emails to the contact identified on our invoice to you. There is a grace period, but, when your API calls are out, we must receive additional payment in order for additional API calls to be serviced.

How many locations are available?


We currently support hundreds of locations from multiple countries around the world. We also nearly fully cover the maps in CARLA. Contact to gain API access to these locations.

What map information does Inverted AI need to add a map I want to add to the back-end?


Inverted AI works internally with the Lanelet2 map format with custom annotations for traffic controls. Contact for more information about how to upload maps to us when requesting new locations to be added to the server side of the API.

How do I add my own map to INITIALIZE or DRIVE?


This is a charged service with no public-facing programmatic interface. Please contact if you are a current customer and would like us to provide INITIALIZE or DRIVE services through our API on your map. Turn-around varies between one to several weeks from initial request to server inclusion.

Is there free academic access to Inverted AI's API products for research purposes?


Not currently. We are able to offer substantially discounted rates to academic institutions however. Please contact and indicate your academic affiliation and needs clearly.

How good are your NPCs?


Our release criteria is that our NPCs jointly drive successfully without any kind of tracked infraction more than 90% of the time in all densities and across a realistic variety of initializations. Often they do much better than that. And they are improving constantly.

What kind of NPCs do you support?


For now, we only support "vehicles" but plans are afoot to explicitly support pedestrians, motorcycles, bicycles, and more soon. Vehicles are distinguished by size in INITIALIZE and DRIVE, meaning that a vehicle the size and shape of a bus is likely to drive like a bus and same for a pickup truck or passenger vehicle.