AirSim — Open Source Simulator for Autonomous Vehicles

Introduction

AirSim is a simulation platform from Microsoft Research for testing autonomous vehicles including drones and cars. Built on Unreal Engine, it provides photorealistic environments with accurate physics, weather simulation, and sensor models for cameras, LiDAR, IMU, and GPS.

What AirSim Does

• Simulates multirotor drones and ground vehicles with realistic aerodynamics and physics
• Provides synthetic sensor data including RGB cameras, depth, segmentation, and LiDAR point clouds
• Supports hardware-in-the-loop (HITL) and software-in-the-loop (SITL) configurations
• Exposes Python and C++ APIs for programmatic control and data collection
• Integrates with reinforcement learning frameworks for training autonomous agents

Architecture Overview

AirSim runs as an Unreal Engine plugin that communicates with external code through an RPC server. The physics engine simulates vehicle dynamics at high frequency while the rendering pipeline generates sensor outputs at configurable rates. Client libraries connect over MessagePack-RPC, allowing control from Python, C++, or any language with an RPC client. Multiple vehicles can run simultaneously in a shared environment.

Self-Hosting & Configuration

• Requires Unreal Engine 4.27+ or pre-built environment binaries for quick start
• Configure vehicles, sensors, and environment via settings.json in the user home directory
• Set camera resolution, FOV, and position offsets per vehicle
• Weather parameters (rain, fog, snow, dust) are adjustable at runtime via API
• Supports Docker deployment for headless training without GPU rendering

Key Features

• Photorealistic rendering with dynamic lighting, shadows, and weather effects
• Domain randomization for robust perception model training
• Built-in recording mode saves synchronized images, poses, and sensor data
• Multi-vehicle support for swarm simulation and cooperative scenarios
• Compatible with PX4, ArduPilot, and custom flight controllers

Comparison with Similar Tools

• Gazebo — better ROS integration but lower visual fidelity; AirSim excels at photorealism
• CARLA — focused on urban driving; AirSim covers both aerial and ground vehicles
• FlightGoggles — MIT drone racing simulator; AirSim offers broader vehicle and sensor support
• Isaac Sim — NVIDIA commercial offering; AirSim is fully open source and free

FAQ

Q: Does AirSim require a powerful GPU? A: For rendering, yes. A modern NVIDIA GPU with 4+ GB VRAM is recommended. Headless mode works without one.

Q: Can I create custom environments? A: Yes, any Unreal Engine environment can host AirSim by adding the plugin.

Q: Is AirSim still maintained? A: Microsoft moved it to maintenance mode in 2022, but the community continues development.

Q: Can I use AirSim for real drone flight testing? A: Yes, HITL mode connects to real flight controllers for pre-flight validation.

Sources

• https://github.com/microsoft/AirSim
• https://microsoft.github.io/AirSim/