Relevant Drone Simulators

This is a quick overview of potentially useful simulators for the project. Please, add any other relevant information.

Colosseum (AirSim)

Colosseum is an active successor to the popular but discontinued Microsoft’s AirSim simulator (Microsoft appears to be trying to commercialize it).

Observations

  • based on Unreal Engine (but can work on Unity to some degree too)
  • it was easy to set up (at least in AirSim days)
  • performant, photorealistic, good physics
  • many sensors supported (multiple cameras, IMU, LIDAR, GPS, etc.)
  • supports many ML use cases
  • weather and illumination can be controlled among other properties
  • quite mature (already years ago)
  • easy-to-use APIs, e.g., in Python and C++
  • support for MavLink and PX4
  • MIT license

Old AirSim Drone Demo Video

Flightmare

UZH’s Fligthmare seems to be rather similar to AirSim/Colosseum at least in terms of photorealism, physics , and available sensor suit. Unlike AirSim, it is based on Unity. The project is also MIT-licensed. However, it’s development seems to be inactive and without any successors.

Flightmare Demo

Gazebo

The Gazebo simulator is also easy-to-setup and work with. The PX4 guide even explains how to setup Gazebo for PX4 software-in-the-loop testing. A drone model similar to our expected drone is even among the default vehicles. It also comes with a permissive, Apache-2.0, license.

However, compared to the previous simulators, the photorealism and performance are lower.

PX4 SITL Gazebo Demo

  • Air Learning: Open-source simulator and a gym environment for deep reinforcement learning research on resource-constrained aerial robots.