AirSim - Open source simulator based on Unreal Engine for autonomous vehicles from Microsoft AI & Research

Jan. 10, 2018, 3:28 p.m. By: Vishakha Jha


AirSim is an Open sourced simulator built on Unreal Engine for autonomous vehicles from Microsoft AI & Research. It is a high-fidelity cross-platform system for testing the safety of AI-based systems. It provides realistic environments, vehicle dynamics and sensing for research. AirSim also works towards physical and visual realistic simulations by extending its support to hardware-in-loop with popular flight controllers such as PX4. It is developed as an Unreal plugin and can be accepted by any environment in a platform independent way with the aim of developing a platform for AI research to experiment with computer vision, deep learning and reinforcement learning algorithms for autonomous vehicles.

It provides users with a variety of features. To choose your vehicle by default AirSim spawns multirotor which allows you to easily switch it to the car and use all the goodies. To manually control the drone remote control (RC) can be used. It exposes APIs through RPC which enables interaction with the vehicle in the simulation programmatically by a variety of languages. These APIs can be really helpful in retrieving images, analysing state, to control the vehicle. They are available as a separate independent cross-platform library which enables you to test your code in the simulator and execute it later on real vehicle providing more flexibility to the user. The Training data can be generated from AirSim for deep learning in two ways either through the record button or by accessing the APIs which provides you full control of how, what, where and when you want to log data. AirSim can also be used in Computer Vision mode which allows you to move around through keyboard and use APIs for positioning of the vehicle.

The Recently updated version aims to expand the functionality of the system to contribute to the advancement of the self-driving vehicle by including car simulation. Other features and improvement include tools for testing airborne vehicles and built-in flight controller that reduces the complexity of setup process. It does not allow expensive debugging and permits controlled rapid experimentation and state estimation algorithms.

Even after achieving this great pace in advancement still building and testing cars in simulation is a very expensive process. It needs a great deal of infrastructure for development of hardware platform, testing and assessment of results. To make various aspects of vehicle automation widely reachable, it is available as an open community-driven platform for testing the algorithms. The new version consists of car simulations, APIs to ease programming, new environments and ready-to-run scripts to speed up the research.

AirSim comes with a very detailed environment that includes a variety of conditions. To ease the testing process of the system, simply drop in the AirSim plugin for simulation that contains more than 12 kilometres of drivable roads spanning more than 20 city blocks. It allows researchers and developers to add new sensors, vehicles or even use different physics engines. It also provides APIs that makes it compatible with a wide variety of languages, including C++ and Python. This makes API easy to use AirSim with various machine learning tool chains. AirSim is available and can be easily downloaded as a compiled binary release. The team has been working towards the next release with new sensors, better vehicle physics, weather modelling and more detailed realistic environments.

More Information : GitHub

AirSim Car Demo:

Video Source: Shital Shah