invertedai 0.0.2

Client SDK for InvertedAI

InvertedAI Drive

Overview

Inverted AI has trained cutting-edge realistic behavioral driving models that are human-like and close the SIM2Real. Our API provides access to these behavioral models and can be useful for several tasks in autonomous vehicle (AV) research and development.

Get Started

In this quickstart tutorial, you’ll run a simple sample AV simulation with Inverted AI Python API. Along the way, you’ll learn key concepts and techniques that are fundamental to using the API for other tasks. In particular, you will be familiar with two main Inverted AI models:

• Drive
• Initialize

Installation

To install use :

pip install invertedai

Setting up

Import the invertedai package and set the API key with add_apikey method.

Refer to the product page to get your API key (or recharge for more tokens).

import invertedai as iai
iai.add_apikey("XXXXXXXXXXXXXX")

Initialize

To run the simulation, the map must be first populated with agents. Inverted AI provides the initialize, a state-of-the-art model trained with real-life driving scenarios which can generate realistic positions for the initial state of the simulation.
Having realistic, complicated and diverse initial conditions are particularly crucial to observer interesting and informative interaction between the agents, i.e., the ego vehicle and NPCs (non-player characters).

You can use initialize in two modes:

• Initialize all agents: generates initial conditions (position and speed) for all the agents including the ego vehicle

response = iai.initialize(
    location="CARLA:Town03:Roundabout",
    agent_count=10,
)

• Initialize NPCs: generates initial conditions (position and speed) only for the NPCs according to the provided state of the ego vehicle.

response = iai.initialize(
    location="CARLA:Town03:Roundabout",
    agent_count=10,
    ego_state=[-11.75, 26.58, 1.36, 4.94],
    ego_attribute=[4.97, 2.04, 1.96]
)

response is a dictionary of states, and agent-attribute (recurrent-states is also returned for compatibility with drive)
response["states"] is a list of agent states, by default the first on the list is always the ego vehicle.

Drive

drive is Inverted AI's cutting-edge realistic driving model trained on millions of miles of traffic data. This model can drive all the agents with only the current state of the environment, i.e., one step observations (which could be obtained from initialize) or with multiple past observations.

response = iai.drive(
    agent_attributes=response["attributes"],
    states=response["states"],
    recurrent_states=response["recurrent_states"],
    get_birdviews=True,
    location="CARLA:Town03:Roundabout",
    steps=1,
)

For convenience and to reduce data overhead, *drive also returns recurrent-states which can be feedbacked to the model instead of providing all the past observations.
Furthermore, drive drive all the agents for $steps\times \frac{1}{FPS}$ where by default $FPS=10[frames/sec]$, should you require other time resolutions contact us.

Running demo locally

Download the examples directory and run:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
.venv/bin/jupyter notebook Drive-Demo.ipynb

Running demo in Colab