• Image annotation

  • LIDAR simulation

  • Image annotation

123

Versatile vehicular software platform

BeamNG.research is a very versatile and open platform for development related to vehicles. Due to its flexibility in its input/output and integrated editors, it is a very good solution to solve problems for the automotive industry and everything around it.

State-of-the-art Soft body physics

We are using an in-house advanced physics engine that enables everyone to build anything out of mass points called ‘nodes’ and connections in between called ‘beams’:

With these basic building blocks, you can build everything:

Example of unintentional behaviors of the physics engine due to the low level simulation:

General tech overview:

Collision detection

Collision detection and resolving at 2000 Hz

Several ground models are available: Asphalt, dirt, gravel, ice, etc

No cheating: Collision resolving is strictly penalty force based, to guarantee physics invariants (momentum conservation etc)

Interoperability

We can import and export data in various ways and formats:

  • Motion/Trajectory/Forces/Telemetry output of various sensors at up to 2k Hz
  • NetworkSocket and Pipe compatible
  • CAN bus (WIP)

  • Example of an android app sending the input and displaying telemetry data:

Simulation

We are simulating a huge variety of subsystems. Some of them:

  • Highly dynamic, interactive drivetrain
  • Brakes: ABS, ECS, thermal simulation
  • High detailed interiors and dynamic dashboards:
  • Engine: Pressure cooled combustion engine
  • Rendering: Good looking visuals:

    Physically based rendering is planned

  • Onboard simulation of devices. For example a simple navigation system with an isolated software:
  • Racing AI that does dynamic path/race line planning without cheating
  • High level scenarios to reenact special scenes

Aerodynamics and hydrodynamics

We are simulating realtime aerodynamics, hydrodynamics and buoyancy:

Pressured volumes:

Pixel perfect image annotation

We can extract images from the simulation:

  • Normal, annotated and depth images (WIP)
  • Good similarity metrics of the extracted images: You can extract a huge amount of different images
  • Our AI drives the vehicles around on our extensive road network in fast-forward
  • Is able to extract a huge amount of images in a short amount of time

Sensors

We are working on various sensors for the vehicles:

  • Forces/Accelerations in high resolution on 2k points in the car
  • Positioning/GPS (WIP)
  • Laser scanners: LIDAR (2.2M points/s at 30 FPS atm)
  • Ultrasound / Distance (WIP)
  • Radar (WIP)
  • Multiple cameras (WIP)
  • Sensor fusion (change sensor setup and re-extract the data) – WIP

Realtime tire physics

In order to calibrate and test our tires, we built an ingame tire tester that ensures the quality of the collision simulation:

More information in our blog: Part 1 / Part 2

Content pipeline

The project is very interoperable with other standards and formats. Some rough keywords:

  • 3D model workflow with 3ds Max/Blender/Maya via Collada (.dae)
  • OpenDrive (WIP)
  • OpenStreetmap (WIP)
  • Integrated level editor:

Autonomous vehicle testing

  • You can extract the data, sensors, images, etc and close the loop back and provide the steering, etc input back to the simulation in realtime.
  • Every vehicle in our simulation is an agent that has its own operating system written in Lua. Changing the behavior is just a matter of a key press, resulting in fast development iteration times

License

  • Free for non-commercial use. We love to participate in research and push the boundaries of what humanity can do together – so it is natural for us to support researchers and their students.
  • Commercial license available. Please contact us.