Simulator List

Choosing a Simulator

< Reference: https://autodrive.readthedocs.io/en/latest/chapters/simulator/comparison.html >

Comparison

../../_images/simulators.gif

Comparison of different simulators:

Table 1: Comparison of Autonomous Simulators
Simulator	Engine	LearningSupport	OpenSource?	Unique Features	Simulator
Deepdrive	Unreal Engine	Tensorflow	YES	+ High FPS due to shared memory+ Terrain is not just flat+ Realistic rendering+ 100GB (8.2 hours) dataset available- Only ground vehicles- Sparse documentation	Deepdrive
Carla	Unreal Engine	TensorFlowChainer	YES	+ Easy to get started- Training code currently unavailable	Carla
Microsoft AirSim	Unreal Engine	CNTK	YES	+ UAV support+ Realtime hardware controller support	Microsoft AirSim
Baidu Apollo	ROSDreamview	TensorflowKeras	YES	+ Support includes free Udacity Apollo course- Hardware focused- Only tested on Lincoln MKZ	Baidu Apollo
Ardupilot	SITL Simulator Gazebo	GymFC	YES	+ UAV support- Hardware focused- Not initially intended for Learning	Ardupilot
TORCS	OpenGL	TensorflowKeras	YES	+ Easy to get started+ Primary research platform for years- Outdated graphics	TORCS
Nvidia DriveWorks	Unreal Engine	Unknown	?	+ Good support for Nvidia GPUs and boards+ Used on Tesla Vehicles	Nvidia DriveWorks
Zoox	Unreal Engine	Unknown	NO	Raised $790,000,000	Zoox
Cruise Automation	Unreal Engine	Unknown	NO	Raised $3,368,800,000	Cruise Automation

Common notes:

• All simulators require dedicated GPU

• Their usage is not always trivial

• SITL: Software in the Loop

• Since our objective is for testing in a simulation environment, hardware focused simulators have been given a negative weightage

Knowlegde prerequisites:

• Linux

• Python

• GitHub

• Docker (Apollo only)

• Machine Learning Practices