AirSim is a high-fidelity visual and physical simulation platform for autonomous vehicles, built on Unreal Engine. It provides realistic simulations for both algorithm development and testing, enabling the generation of large amounts of annotated training data for machine learning. The simulator includes a physics engine that supports high-frequency real-time hardware-in-the-loop (HITL) simulations with protocols like MavLink. It is designed to be extensible, allowing for new vehicle types, hardware platforms, and software protocols. The modular design enables components to be used independently in other projects. The simulator is used to test autonomous vehicles, such as quadrotors, by comparing simulated software components with real-world flights. It addresses challenges in training autonomous systems by providing realistic environments and sensor models. The platform aims to bridge the gap between simulation and reality, supporting the development of data-driven machine intelligence techniques like reinforcement learning and deep learning. AirSim's architecture includes environment models, vehicle models, physics engines, sensor models, rendering interfaces, and APIs. The vehicle model defines a rigid body with parameters for mass, inertia, and drag. The environment includes gravity, air pressure, and magnetic field models. The physics engine computes the next kinematic state of bodies in the simulated world, considering forces and torques from actuators. Sensors in AirSim include barometers, gyroscopes, accelerometers, magnetometers, and GPS. These sensors are modeled with realistic noise and drift, using data from sensor datasheets. The visual rendering is based on Unreal Engine 4, providing high-fidelity environments and realistic lighting and shadows. Experiments show that AirSim's simulation closely matches real-world performance, with small differences due to factors like integration errors and model approximations. The simulator's sensor models, including barometers, magnetometers, and IMUs, closely match real-world devices. Future work includes improving collision response, ground interaction models, and adding advanced noise and lens models for cameras, as well as simulating GPS signal degradation and wind effects. AirSim is designed to be extensible, supporting future vehicle types and applications.AirSim is a high-fidelity visual and physical simulation platform for autonomous vehicles, built on Unreal Engine. It provides realistic simulations for both algorithm development and testing, enabling the generation of large amounts of annotated training data for machine learning. The simulator includes a physics engine that supports high-frequency real-time hardware-in-the-loop (HITL) simulations with protocols like MavLink. It is designed to be extensible, allowing for new vehicle types, hardware platforms, and software protocols. The modular design enables components to be used independently in other projects. The simulator is used to test autonomous vehicles, such as quadrotors, by comparing simulated software components with real-world flights. It addresses challenges in training autonomous systems by providing realistic environments and sensor models. The platform aims to bridge the gap between simulation and reality, supporting the development of data-driven machine intelligence techniques like reinforcement learning and deep learning. AirSim's architecture includes environment models, vehicle models, physics engines, sensor models, rendering interfaces, and APIs. The vehicle model defines a rigid body with parameters for mass, inertia, and drag. The environment includes gravity, air pressure, and magnetic field models. The physics engine computes the next kinematic state of bodies in the simulated world, considering forces and torques from actuators. Sensors in AirSim include barometers, gyroscopes, accelerometers, magnetometers, and GPS. These sensors are modeled with realistic noise and drift, using data from sensor datasheets. The visual rendering is based on Unreal Engine 4, providing high-fidelity environments and realistic lighting and shadows. Experiments show that AirSim's simulation closely matches real-world performance, with small differences due to factors like integration errors and model approximations. The simulator's sensor models, including barometers, magnetometers, and IMUs, closely match real-world devices. Future work includes improving collision response, ground interaction models, and adding advanced noise and lens models for cameras, as well as simulating GPS signal degradation and wind effects. AirSim is designed to be extensible, supporting future vehicle types and applications.