AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles

Shital ShahDebadeepta DeyChris LovettAshish Kapoor

article2017FSR2,542 citations

Presents AirSim, an Unreal Engine-based simulator that combines high-frequency physics and photorealistic environments to enable real-time hardware-in-the-loop testing and large-scale data generation for autonomous vehicles.

Listen

The article addresses the high cost and time required to develop and test autonomous vehicle algorithms in the real world, along with the difficulty of obtaining large volumes of annotated training data needed for machine learning methods such as reinforcement learning. These challenges are compounded by the gap between simplified simulations and real-world conditions, which hinders effective transfer of learned behaviors.

The article set out to create and validate an open-source simulator that delivers both physically accurate dynamics and visually rich environments to support algorithm development, data generation, and hardware-in-the-loop testing for autonomous vehicles.

The authors built AirSim as a modular plugin for Unreal Engine 4, incorporating a high-frequency physics engine, detailed vehicle and environment models, realistic sensor simulations, and APIs that allow the same code to run in simulation or on real hardware. They implemented a quadrotor model and conducted controlled experiments comparing simulated flights against real-world flights using identical commands and hardware parameters.

The simulation reproduced real flight trajectories with Hausdorff distances of 0.65 m for square paths and 1.47 m for circular paths. Barometer and magnetometer outputs closely matched real sensor traces, while IMU variance aligned well during stationary conditions but underestimated vibration effects during flight.

These results indicate that AirSim can generate large quantities of realistic training data at low cost and enable algorithms developed in simulation to transfer directly to physical vehicles, reducing both development expense and safety risks. The findings matter because they lower barriers to testing data-driven autonomy techniques that would otherwise require extensive real-world operation.

Further work is needed to model airframe vibrations, advanced collision responses, GPS signal degradation, and richer wind effects before broader deployment decisions. The main limitations are the use of simplified models for real-time performance and incomplete capture of certain dynamic phenomena, so users should validate critical behaviors on hardware before relying on simulation results for final certification or high-risk applications.

Microsoft deep learning engine hardware platform autonomous vehicles physical simulation physics engine

arXiv: 1705.05065 microsoft/AirSim

Paper: The SYNTHIA Dataset: A Large Collection of Synthetic Images for Semantic Segmentation of Urban Scenes, Germán Ros et al. (2016). Demonstrates the foundational utility of synthetic urban visual data for training autonomous vehicle perception models before AirSim unified visual and dynamic simulation.
Paper: OpenAI Gym, Greg Brockman et al. (2016). Establishes the standardized agent-environment interface paradigm for reinforcement learning that modern high-fidelity simulators adapt for embodied autonomy.
Paper: Domain randomization for transferring deep neural networks from simulation to the real world, Josh Tobin et al. (2017). Introduces domain randomization as a primary methodology for bridging the simulation-to-reality gap in vision-based robotics.
Paper: A Survey of Motion Planning and Control Techniques for Self-Driving Urban Vehicles, Brian Paden et al. (2016). Provides a comprehensive foundation of motion planning and vehicle dynamic control models essential for building realistic autonomous vehicle simulators.
Paper: A comprehensive survey on safe reinforcement learning, Javier García et al. (2015). Surveys the hazards and constraints of physical trial-and-error exploration, highlighting the critical safety motivations behind developing high-fidelity simulators.

Paper: CARLA: An Open Urban Driving Simulator, Alexey Dosovitskiy et al. (2017). Extends the Unreal Engine-based simulation paradigm specifically to urban driving with integrated traffic systems and multi-agent benchmarking.
Paper: Deep Reinforcement Learning for Autonomous Driving: A Survey, B Ravi Kiran et al. (2020). Synthesizes how high-fidelity simulation environments like AirSim enable deep reinforcement learning paradigms for autonomous driving while addressing sim-to-real transfer.
Paper: Habitat: A Platform for Embodied AI Research, Manolis Savva et al. (2019). Develops high-throughput, photorealistic 3D simulation platforms tailored for scaling up embodied AI training and navigation.
Paper: Open3D: A Modern Library for 3D Data Processing, Qian-Yi Zhou et al. (2018). Provides an open-source library for processing and evaluating 3D sensor point clouds and meshes commonly generated by spatial simulation pipelines.
Paper: Learning dexterous in-hand manipulation, Marcin Andrychowicz et al. (2018). Applies large-scale simulation and physical domain randomization to transfer complex manipulation policies directly to real robotic hardware.

Cover for AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles

Abstract

Developing and testing algorithms for autonomous vehicles in real world is an expensive and time consuming process. Also, in order to utilize recent advances in machine intelligence and deep learning we need to collect a large amount of annotated training data in a variety of conditions and environments. We present a new simulator built on Unreal Engine that offers physically and visually realistic simulations for both of these goals. Our simulator includes a physics engine that can operate at a high frequency for real-time hardware-in-the-loop (HITL) simulations with support for popular protocols (e.g. MavLink). The simulator is designed from the ground up to be extensible to accommodate new types of vehicles, hardware platforms and software protocols. In addition, the modular design enables various components to be easily usable independently in other projects. We demonstrate the simulator by first implementing a quadrotor as an autonomous vehicle and then experimentally comparing the software components with real-world flights.

Topics

deep learning engine hardware platform self-driving auto software component hard-ware-in-the-loop autonomous robot

Keywords

autonomous vehicles physical simulation physics engine flight controller Unreal engine sensor models computer vision collision response kinematic state

1 Introduction
2 Related Work
3 Architecture
3.1 Vehicle Model
3.2 Environment
Gravity
Magnetic Field
Air Pressure and Density
3.3 Physics Engine
Linear and Angular drag
Accelerations
Integration
Collisions
3.4 Sensors
Barometer
Gyroscope and Accelerometer
Magnetometer
Global Positioning System (GPS)
3.5 Visual Rendering
4 Experiments
5 Conclusion and Future Work
References

Knowls

Knowl 1 — AirSim Simulator Architecture
model/method
AirSim is an open-source simulation platform designed as a plugin for Unreal Engine 4 (UE4) to simulate autonomous vehicles with high physical and visual fidelity.

The system is organized into modular, independently reusable components:
1. Environment Model: Computes ambient physical properties such as altitude-dependent gravity, atmospheric pressure, air density, and the geomagnetic field.
2. Vehicle Model: Represents the vehicle as a multi-actuator rigid body and converts actuator control commands into applied forces and torques.
3. Physics Engine: Integrates kinematics at a high update rate ( $1000\text{ Hz}$ ) to support real-time high-speed flight dynamics and hardware-in-the-loop (HITL) execution.
4. Sensor Models: Implements modular, header-only C++ sensor abstractions (IMU, barometer, magnetometer, GPS) with realistic noise and drift.
5. Rendering Interface: Leverages UE4's rendering capabilities (physically based materials, photometric lighting, ray-traced distance field shadows, reflections) and collision mesh geometry.
6. Vehicle Firmware Interface: Bridges the simulation with vehicle flight controller firmware (such as PX4, ROSFlight, and Hackflight) over standard protocols including MavLink in both HITL and software-in-the-loop (SITL) modes.
7. Public API Layer: Exposes sensor streams, kinematic states, and control commands to companion computers (e.g., executing SLAM or path planning). The API abstracts whether the algorithm runs in simulation or on a physical vehicle, allowing direct deployment of control code to real platforms.
Knowl 2 — AirSim Multi-Actuator Vehicle Dynamics Model
model/method
AirSim models an autonomous vehicle as a rigid body defined by its total mass $m$ , inertia tensor $I$ , linear and angular drag coefficients, surface cross-sectional area $A$ , and coefficients of friction and restitution.

The vehicle is actuated through a set of $K$ vertices located at body-frame positions $\{r_1, \dots, r_K\}$ with surface unit normal vectors $\{n_1, \dots, n_K\}$ . Each vertex $i$ receives a dimensionless scalar control input $u_i \in [0, 1]$ . The positions $r_i$ and normals $n_i$ can dynamically vary during flight to model configurations such as Vertical Take-Off and Landing (VTOL) and morphing airframes.

For a quadrotor ( $K=4$ ), the thrust force $F_i$ and torque $\tau_i$ generated at propeller vertex $i$ are computed as:

$F_i = C_T \rho \omega_{\max}^2 D^4 u_i \cdot n_i$

$\tau_i = \frac{1}{2\pi} C_{\text{pow}} \rho \omega_{\max}^2 D^5 u_i \cdot n_i$

where:
- $C_T$ is the dimensionless propeller thrust coefficient.
- $C_{\text{pow}}$ is the dimensionless propeller power coefficient.
- $\rho$ is the ambient air density in $\text{kg/m}^3$ .
- $D$ is the propeller diameter in meters.
- $\omega_{\max}$ is the maximum propeller angular velocity in revolutions per minute (RPM).
Knowl 3 — AirSim High-Frequency Physics Engine and Kinematic Integration
algorithm
The AirSim physics engine evaluates rigid-body dynamics at an update rate of $1000\text{ Hz}$ . Translational kinematics are integrated using the second-order Velocity Verlet algorithm, while rotational kinematics are maintained in body-frame angular velocity and integrated into world-frame orientation quaternions.

Linear drag force $F_d$ acts opposite to the velocity vector $v$ with magnitude:

$|F_d| = \frac{1}{2} \rho |v|^2 C_{\text{lin}} A$

where $C_{\text{lin}}$ is the linear drag coefficient, $A$ is the cross-sectional area, and $\rho$ is air density. Angular drag torque $\tau_d$ is calculated by integrating infinitesimal surface drag forces $dF = -\frac{1}{2} \rho |r_{ds} \times \omega| (r_{ds} \times \omega) C_{\text{lin}} ds$ over the vehicle surface (approximated as a rectangular box):

$\tau_d = \int_S r_{ds} \times dF$
Input: State at step $k$ : position $p_k$ , orientation quaternion $q_k$ , linear velocity $v_k$ , angular velocity $\omega_k$ , linear acceleration $a_k$
Input: Actuator forces $\{F_i\}$ and torques $\{\tau_i\}$ at body offsets $\{r_i\}$
Input: Mass $m$ , inertia tensor $I$ , air density $\rho$ , linear drag coefficient $C_{\text{lin}}$ , cross-section $A$ , gravity $g$ , time step $dt$
Output: State at step $k+1$ : $p_{k+1}, q_{k+1}, v_{k+1}, \omega_{k+1}, a_{k+1}, \alpha_{k+1}$
Compute linear drag force: $F_d = -\frac{1}{2} \rho |v_k| v_k C_{\text{lin}} A$
Compute angular drag torque $\tau_d = \int_S r_{ds} \times dF$
Compute net translational force: $F_{\text{net}} = \sum_i F_i + F_d$
Compute net torque in body frame: $\tau_{\text{net}} = \sum_i (\tau_i + r_i \times F_i) + \tau_d$
Compute translational acceleration: $a_{k+1} = \frac{F_{\text{net}}}{m} + g$
Compute body angular acceleration: $\alpha_{k+1} = I^{-1} (\tau_{\text{net}} - (\omega_k \times (I \cdot \omega_k)))$
Update linear velocity (Velocity Verlet): $v_{k+1} = v_k + \frac{a_k + a_{k+1}}{2} dt$
Update position (Velocity Verlet): $p_{k+1} = p_k + v_k dt + \frac{1}{2} a_k dt^2$
Update angular velocity: $\omega_{k+1} = \omega_k + \alpha_{k+1} dt$
Compute rotation angle: $\alpha_{dt} = |\omega_{k+1}| dt$
Compute rotation unit axis: $u = \frac{\omega_{k+1}}{|\omega_{k+1}|}$ if $|\omega_{k+1}| > 0$ else $[0, 0, 0]^T$
Construct body rotation quaternion: $q_{dt} = [\cos(\alpha_{dt}/2), u \sin(\alpha_{dt}/2)]$
Update world orientation quaternion: $q_{k+1} = q_k \cdot q_{dt}$
return $p_{k+1}, q_{k+1}, v_{k+1}, \omega_{k+1}, a_{k+1}, \alpha_{k+1}$
Knowl 4 — AirSim Atmospheric Pressure and Gravitational Field Models
model/method
AirSim models variation in gravity, atmospheric temperature, atmospheric pressure, and air density as functions of altitude $h$ above sea level:
1. Gravity Model: Gravitational acceleration $g$ at height $h$ is approximated using a first-order binomial expansion of Newton's law of universal gravitation:
$g = g_0 \frac{R_e^2}{(R_e + h)^2} \approx g_0 \left(1 - 2\frac{h}{R_e}\right)$

where $g_0$ is the standard surface gravitational constant ( $9.80665\text{ m/s}^2$ ) and $R_e$ is the mean Earth radius.
1. Atmospheric Temperature and Pressure: Standard Temperature $T$ and Standard Pressure $P$ are computed using the 1976 U.S. Standard Atmosphere model for altitudes below $51\text{ km}$ , and extended up to $86\text{ km}$ using standard atmospheric tables.
2. Air Density: Density $\rho$ is computed from standard pressure $P$ and absolute temperature $T$ using the ideal gas law:
$\rho = \frac{P}{R \cdot T}$

where $R = 287.058\text{ J/(kg}\cdot\text{K)}$ is the specific gas constant for dry air.
Knowl 5 — Tilted Dipole Geomagnetic Field Model in AirSim
model/method
To model Earth's magnetic field in real time without the computational cost of high-degree spherical harmonic models like WMM2015, AirSim implements a first-order tilted dipole model assuming Earth is a perfect dipole sphere.

Given the vehicle's latitude $\theta$ , longitude $\phi$ , altitude $h$ , and the magnetic north pole position $(\theta^0, \phi^0)$ :
1. The magnetic co-latitude $\theta_m$ is determined by:
$\cos\theta_m = \cos\theta \cos\theta^0 + \sin\theta \sin\theta^0 \cos(\phi - \phi^0)$
1. The total magnetic field intensity $|B|$ is calculated as:
$|B| = B_0 \left(\frac{R_e}{R_e + h}\right)^3 \sqrt{1 + 3\cos^2\theta_m}$

where $B_0$ is the mean magnetic field strength at the geomagnetic equator and $R_e$ is Earth's mean radius.
1. The magnetic inclination angle $\alpha$ and declination angle $\beta$ are given by:
$\tan\alpha = 2\cot\theta_m$

$\sin\beta = \begin{cases} \frac{\sin(\phi - \phi^0)\cos\theta^0}{\sin\theta_m}, & \text{if } \cos\theta_m > \sin\theta^0 \sin\theta \\ \frac{\cos(\phi - \phi^0)\cos\theta^0}{\sin\theta_m}, & \text{otherwise} \end{cases}$
1. The horizontal field intensity $H$ , latitudinal component $X$ , longitudinal component $Y$ , and vertical downward component $Z$ are resolved as:
$H = |B|\cos\alpha, \quad Z = |B|\sin\alpha, \quad X = H\cos\beta, \quad Y = H\sin\beta$
Knowl 6 — AirSim Inertial Measurement Unit (IMU) Sensor Noise Model
model/method
AirSim simulates 3-axis gyroscopes and accelerometers by superimposing white Gaussian noise and a time-varying random-walk bias drift onto ground-truth kinematic values.

Given the ground-truth angular velocity in the body frame $\omega$ , the simulated gyroscope output $\omega^{\text{out}}$ is:

$\omega^{\text{out}} = \omega + \eta_a + b_t$

$\eta_a \sim \mathcal{N}(0, r_a)$

$b_t = b_{t-1} + \eta_b, \quad \eta_b \sim \mathcal{N}\left(0, b_0 \sqrt{\frac{dt}{t_a}}\right)$

where:
- $r_a$ is the white noise variance.
- $b_0$ is the bias parameter.
- $t_a$ is the bias drift time constant.
- $dt$ is the discrete update step.
For the accelerometer, the simulator first subtracts the gravitational acceleration vector from the true linear acceleration in the world frame ( $a_{\text{world}} - g$ ), transforms the resulting specific acceleration into the vehicle body frame, and then adds Gaussian white noise and random-walk bias drift parameterized in the same form as the gyroscope.
Knowl 7 — AirSim Barometric Altimeter Sensor Model
model/method
AirSim simulates a barometric altimeter by querying the ground-truth standard atmospheric pressure $P$ at the vehicle's altitude and applying a first-order Gauss-Markov bias drift process combined with high-frequency Gaussian measurement noise.

The time-dependent bias factor $b_k$ is updated as:

$b_{k+1} = w \cdot b_k + (1 - w) \cdot \eta$

$w = \exp\left(-\frac{dt}{\tau}\right), \quad \eta \sim \mathcal{N}(0, s^2)$

where $\tau = 1\text{ hour}$ is the process time constant and $s$ is the drift standard deviation. Sensor measurement noise is drawn from $\mathcal{N}(0, \sigma^2)$ using manufacturer datasheet parameters (e.g., MEAS MS5611-01BA).

The simulated pressure measurement $p$ is converted to indicated altitude $h$ using the standard barometric formula:

$h = \frac{T_0}{a} \left[ \left(\frac{p}{p_0}\right)^{-\frac{a \cdot R}{g}} - 1 \right]$

where $T_0 = 288.15\text{ K}$ ( $15^\circ\text{C}$ ), $a = -6.5 \times 10^{-3}\text{ K/m}$ is the vertical temperature gradient, $R = 287.058\text{ J/(kg}\cdot\text{K)}$ is the specific gas constant, $g$ is gravitational acceleration, and $p_0$ is sea-level reference pressure.
Knowl 8 — AirSim GPS and Magnetometer Sensor Models
model/method
AirSim provides abstract sensor models for magnetometers and Global Positioning System (GPS) receivers:
1. Magnetometer Model: Computes ground-truth magnetic field vector components $(X, Y, Z)$ in the vehicle body frame from geographic coordinates using the tilted dipole model, and adds zero-mean additive white Gaussian noise specified by the sensor datasheet (e.g., Honeywell HMC5883).
2. GPS Model:
- Simulates transmission and processing latency (configured to a typical default of $200\text{ ms}$ ).
- Simulates lower update rates (configured to a typical default of $50\text{ Hz}$ ).
- Models satellite fix acquisition over time by decaying horizontal and vertical position error estimates via individually parameterized first-order low-pass filters.
Knowl 9 — Real-World Flight Trajectory Tracking Validation of AirSim
empirical result
AirSim's flight dynamics were evaluated against physical flights using a DJI Flamewheel quadrotor equipped with a Pixhawk v2 flight controller and an onboard Gigabyte 5500 Brix computer running Ubuntu 16.04. The AirSim quadrotor model was configured using the physical dimensions, mass parameters, and sensor datasheet specifications of the physical vehicle. Identical offboard control commands were dispatched via the AirSim MavLinkTest application.

Two geometric patterns were executed and compared in local North-East-Down (NED) coordinates:
1. Square Trajectory ( $5\text{ m}$ side length): The symmetric Hausdorff distance between the physical quadrotor track and the simulated track was $0.65\text{ m}$ .
2. Circular Trajectory ( $10\text{ m}$ radius): The symmetric Hausdorff distance between the physical quadrotor track and the simulated track was $1.47\text{ m}$ .
Discrepancies between real and simulated flights are attributed to numerical integration approximations, rigid-body approximations of the airframe, and unmodeled real-world wind gusts.
Knowl 10 — Sensor Accuracy Comparison and Motor Vibration IMU Limitation
empirical result
Individual sensor models in AirSim were evaluated against physical sensor hardware configured via manufacturer datasheets:
1. Barometer (MEAS MS5611-01BA): Raising the sensor periodically between ground level and $178\text{ cm}$ produced matching dynamic pressure waveforms, with minor offsets caused by ambient temperature changes and local atmospheric differences.
2. Magnetometer (Honeywell HMC5883): Rotating the platform on the ground in four sequential $90^\circ$ steps produced closely matching $X$ and $Y$ Gauss field measurements between simulation and reality.
3. IMU (InvenSense MPU 6000):
- Stationary Condition:
  - Gyroscope variance: Simulated $= 2.47 \times 10^{-7}\text{ rad}^2/\text{s}^2$ ; Real-world $= 6.71 \times 10^{-7}\text{ rad}^2/\text{s}^2$ .
  - Accelerometer variance: Simulated $= 1.78 \times 10^{-4}\text{ m}^2/\text{s}^4$ ; Real-world $= 1.93 \times 10^{-4}\text{ m}^2/\text{s}^4$ .
- In-Flight Condition:
  - Accelerometer variance: Simulated $= 1.75 \times 10^{-3}\text{ m}^2/\text{s}^4$ ; Real-world $= 9.46\text{ m}^2/\text{s}^4$ .
Limitation: AirSim's baseline physics engine does not model high-frequency motor-induced airframe vibrations, causing simulated in-flight accelerometer variance to be substantially lower than real-world flight measurements.

Coverage note — No substantial contributed material was omitted; all core architectural components, physical and environmental models, sensor formulations, numerical integration methods, and empirical validation benchmarks are represented.

References

1.Babushkin, A.: Jmavsim. https://pixhawk.org/dev/hil/jmavsim
2.Bagnell, J.A.: An invitation to imitation. Tech. rep., CMU ROBOTICS INST (2015)
3.Braeunig, R.: Atmospheric models. http://www.braeunig.us/space/atmmodel.htm (2014)
4.Brandt, J., Deters, R., Ananda, G., Selig, M.: Uiuc propeller database, university of illinois at urbana-champaign. http://m-selig.ae.illinois.edu/props/propDB.html (2015)
5.Burch D., B.T.: Mariner’s Pressure Atlas: Worldwide Mean Sea Level Pressures and Standard Deviations for Weather Analysis. Starpath School of Navigation (2014)
6.Chulliat, A., Macmillan, S., Alken, P., Beggan, C., Nair, M., Hamilton, B., Woods, A., Ridley, V., Maus, S., Thomson, A.: The us/uk world magnetic model for 2015-2020 (2015). DOI 10.7289/V5TB14V7
7.Furrer, F., Burri, M., Achtelik, M., Siegwart, R.: Rotorsa modular gazebo mav simulator framework. In: Robot Operating System (ROS), pp. 595–625. Springer (2016)
8.Hecker, C.: Physics, part 3: Collision response. Game Developer Magazine (1997)
9.Herman, R.: A first course in differential equations for scientists and engineers. http://people.uncw.edu/hermanr/mat361/ODEBook/ (2017)
10.Jackson, J., Ellingson, G., McLain, T.: Rosflight: A lightweight, inexpensive mav research and development tool. In: ICUAS, pp. 758–762 (2016). DOI 10.1109/ICUAS.2016.7502584
11.Karis, B., Games, E.: Real shading in unreal engine 4. In: Proc. Physically Based Shading Theory Practice (2013)
12.Kober, J., Bagnell, J.A., Peters, J.: Reinforcement learning in robotics: A survey. Int. J. Rob. Res. 32(11), 1238–1274 (2013). DOI 10.1177/0278364913495721
13.Koenig, N., Howard, A.: Design and use paradigms for gazebo, an open-source multi-robot simulator. In: IROS (2004)
14.Lanza, R., Meloni, A.: The Earth’s Magnetism: An Introduction for Geologists. Springer Science & Business Media (2006)
15.Levy, S.: Hackflight: Simple quadcopter flight control firmware and simulator for c++ hackers. https://github.com/simondlevy/hackflight
16.Meier, L., Tanskanen, P., Fraundorfer, F., Pollefeys, M.: Pixhawk: A system for autonomous flight using onboard computer vision. In: ICRA, pp. 2992–2997. IEEE (2011)
17.Meyer, J., Sendobry, A., Kohlbrecher, S., Klingauf, U., Von Stryk, O.: Comprehensive simulation of quadrotor uavs using ros and gazebo. In: SIMPAR, pp. 400–411. Springer (2012)
18.Moore, H.: Creating assets for the open world demo (2015)
19.Nakayama, Y., Boucher, R.: Introduction to fluid mechanics. Butterworth-Heinemann (1998)
20.Sabatini, A.M., Genovese, V.: A stochastic approach to noise modeling for barometric altimeters. Sensors (Basel, Switzerland) 13(11), 15,692–15,707 (2013)
21.Shah, S., Dey, D., Lovett, C., Kapoor, A.: Airsim open source platform at github. https://github.com/Microsoft/AirSim (2017)
22.Stull, R.: Practical Meteorology: An Algebra-based Survey of Atmospheric Science. University of British Columbia (2015)
23.Tapley, B., Ries, J., Bettadpur, S., Chambers, D., Cheng, M., Condi, F., Poole, S.: The ggm03 mean earth gravity model from grace. In: American Geophysical Union, G42A-03 (2007)
24.Taylor, J.: Classical mechanics. University Science Books (2005)
25.Weiss, K., Khoshgoftaar, T.M., Wang, D.: A survey of transfer learning. Journal of Big Data 3(1), 9 (2016). DOI 10.1186/s40537-016-0043-6
26.Woodman, O.J.: An introduction to inertial navigation. Tech. Rep. UCAM-CL-TR-696, University of Cambridge, Computer Laboratory (2007)

Citation

MLA

Shah, S., et al. “AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles”. Springer Proceedings in Advanced Robotics, Springer International Publishing, 2017, pp. 621–35, https://doi.org/10.1007/978-3-319-67361-5_40.

APA

Shah, S., Dey, D., Lovett, C., & Kapoor, A. (2017). AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles. In Springer Proceedings in Advanced Robotics (pp. 621–635). Springer International Publishing. https://doi.org/10.1007/978-3-319-67361-5_40

Chicago

Shah, S., D. Dey, C. Lovett, and A. Kapoor. 2017. “AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles”. In Springer Proceedings in Advanced Robotics. Springer International Publishing. https://doi.org/10.1007/978-3-319-67361-5_40.

Harvard

Shah, S. et al. (2017) “AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles”, Springer Proceedings in Advanced Robotics. Springer International Publishing, pp. 621–635. Available at: https://doi.org/10.1007/978-3-319-67361-5_40.

Vancouver

1. Shah S, Dey D, Lovett C, Kapoor A (2017) AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles. In: Springer Proceedings in Advanced Robotics. Springer International Publishing, pp 621–635

BibTeX

@inbook{Shah_2017, title={AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles}, ISBN={9783319673615}, ISSN={2511-1264}, url={http://dx.doi.org/10.1007/978-3-319-67361-5_40}, DOI={10.1007/978-3-319-67361-5_40}, booktitle={Field and Service Robotics}, publisher={Springer International Publishing}, author={Shah, Shital and Dey, Debadeepta and Lovett, Chris and Kapoor, Ashish}, year={2017}, month=Nov, pages={621–635} }

Metadata:Crossref

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

AirSim: High-Fidelity Visual and Physical Simulation for Autonomous Vehicles

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — AirSim Simulator Architecture

Knowl 2 — AirSim Multi-Actuator Vehicle Dynamics Model

Knowl 3 — AirSim High-Frequency Physics Engine and Kinematic Integration

Knowl 4 — AirSim Atmospheric Pressure and Gravitational Field Models

Knowl 5 — Tilted Dipole Geomagnetic Field Model in AirSim

Knowl 6 — AirSim Inertial Measurement Unit (IMU) Sensor Noise Model

Knowl 7 — AirSim Barometric Altimeter Sensor Model

Knowl 8 — AirSim GPS and Magnetometer Sensor Models

Knowl 9 — Real-World Flight Trajectory Tracking Validation of AirSim

Knowl 10 — Sensor Accuracy Comparison and Motor Vibration IMU Limitation

References

Citation

BibTeX

Source Code

Access the Paper