Implementation of a HITL-Enabled High Autonomy Drone Architecture on a Photo-Realistic Simulator

Abstract

In the development process of Unmanned Aerial Vehicles (UAVs), researchers have to solve a lot of complex problems such as unreliable pose estimation, sensor errors, unknown and fast-changing environments, which require a lot of test trials. However, real-world testing of UAVs is expensive, time-consuming, and requires a pilot and careful testing of the correct integration of the hardware and software components. As a result, the usage of simulation tools presents itself as a promising alternative to real flight tests. Therefore, having a nearly photo-realistic, Hardware in the Loop (HITL) enabled and accurate simulation tool integrated with a popular autopilot board would be advantageous to the research community. We propose an implementation of a high autonomy drone navigation architecture integrated with the photo-realistic environment simulator Flightmare Renderer, the flight-dynamics simulator Gazebo and a popular autopilot board Pixhawk (PX4) controller to make a photo-realistic and accurate dynamic simulation for drones. By means of our simulation, we evaluate the drone system architecture by running a full set of algorithms required to achieve full autonomy, which includes visual-inertial pose estimation, mapping, path planning, obstacle avoidance, and control modules. That means each individual module and the whole system are tested and evaluated on the photo-realistic rendering and accurate flight-dynamics simulation environment. Thanks to the integration of the PX4 autopilot with HITL simulation, the progress from simulation flight to real flight experiments is easier and faster.