It's impossible to exhaust search the real-world environment and generate enough dataset to train a robot to perceive the world, there comes to simulators.
Our goal is to build a pipeline that can generate a indoor scene simulator that a robot can roam freely and, at the same time, it can yield the randomized scenario (random light condition, random color, random permutation of facilities and cluttered objects) on-the-fly.
Luckily @Microsoft and @UnrealEngine has done 99% of the work for us. AirSim is such a plugin that can generate most of the perception necessary data from a custom scene.
What we need to do is simply:
- download AirSim and follow its tutorial to setup the system on Windows
- download EpicGames and install
UnrealEngine-4.25from its marketplace - download custom scene (Singapore Condominium) from here with extraction code
33n8
- setup and build the custom environment to support
AirSim
when you see your custom scene has the following gui interface it means you have successfully built it:
!!note!!: you might need to press 1 2 3 key to activate these interface, please read the documentation of AirSim carefully, since they might change the key bindings.
bonus point:
- add random day-time sunlight effect
- add a moving robot model to the custom scene
- setup the ros2 bridge and record a rosbag
©midea_lab2030