rlPx4Controller (https://github.com/emNavi/rlPx4Controller) is a quadcopter control library implemented in C++ and provides a python interface through pybind11. The motivation of this work is to build a parallel flight control for large-scale quadrotor reinforcement learning (RL) and Sim-to-Real on the PX4 stack. Thus, the implementation of the controller is consistent with the Px4 flight control to ensure the consistency of sim2real.

Features

• CPU-based parallel flight geometric control computation.
• Strict PX4-like flight controller with differential flatness.
• Parallel control for different levels that supports: position & yaw control (PY), velocity & yaw control (LV), collective thrust & attitude angle (CTA), collective thrust & body rate (CTBR)

4 modules have been implemented so far:

• PolyTrajGen: Fifth degree polynomial trajectory generation
• Lemniscate: Figure-eight curve generation
• pyControl: Single aircraft controller
• pyParallelControl：Multi-aircraft controller for reinforcement learning in vectorized environments.

Usage

rlPx4Controller provides a very simple way to install and use:

Install:

git clone git@github.com:emNavi/rlPx4Controller.git
cd rlPx4Controller
pip install -e .

Use:

from rlPx4Controller import ParallelRateControl, ParallelVelControl, ParallelAttiControl, ParallelPosControl
# According to which control layer is selected
# Use Rate control as an example
controller = ParallelRateControl(num_envs)
controller.set_status(status, dt)
cmd_thrusts = controller.update(actions)

num_envs denotes how many environments are paralleled, dt is controller's simulation interval. set_status() inputs the status of the drone, including position, attitude in quaternion, linear velocity, and angular velocity. update() returns commands of thrust.