Freestyle Quadcopter Flight in Pybullet with Gym and (soon) PettingZoo APIs
Project description
PyFlyt - Freestyle Quadcopter Flight in Pybullet with Gymnasium and (soon) PettingZoo APIs
This is a library for running reinforcement learning algorithms on both real crazyflies and simulated ones using the Gymnasium and (soon) PettingZoo APIs.
This repo's master branch is still under development.
Inspired by the original pybullet drones by University of Toronto's Dynamic Systems Lab with several key differences:
- Actual full cascaded PID flight controller implementations for each drone.
- Actual motor RPM simulation using first order differential equation.
- More modular control structure
- For developers - 8 implemented flight modes that use tuned cascaded PID flight controllers, available in
PyFlyt/core/drone.py. - For developers - easily build your own multiagent environments using the
PyFlyt.core.aviary.Aviaryclass. - More environments with increasing difficulties, targetted at enabling hiearchical learning for as true-to-realistic freestyle quadcopter flight.
Table of Contents
Installation
pip3 install wheel
pip3 install pyflyt
Usage
Usage is similar to any other Gymnasium and (soon) PettingZoo environment:
import gymnasium
import PyFlyt
env = gymnasium.make("PyFlyt/SimpleHoverEnv-v0")
# ommit the below line to remove renderring and let
# the simulation go as fast as possible
env.render()
obs = env.reset()
done = False
while not done:
observation, reward, done, _ = env.step(env.observation_space.sample())
Observation Space
All observation spaces use gymnasium.spaces.Box.
For Simple environments, the observation spaces are simple 1D vectors of length less than 25, all values are not normalized.
For Advanced environments, the observation spaces use gymnasium.spaces.Dict with an image and state component.
Action Space
All environments use the same action space, this is to allow hiearchical learning to take place - an RL agent can learn to hover before learning to move around.
By default, all environments have an action space that corresponds to FPV controls - pitch angular rate, roll angular rate, yaw_angular rate, thrust. The limits for angular rate are +-3 rad/s. The limits for thrust commands are -1 to 1.
The angular rates are intentionally not normalized to allow for a) better interpretation and b) more realistic inputs.
Environments
PyFlyt/SimpleHoverEnv-v0
A simple environment where an agent can learn to hover. The environment ends when either the Quadcopter collides with the ground or exits the permitted flight dome.
PyFlyt/SimpleWaypointEnv-v0
A simple environment where the goal is to position the Quadcopter at random setpoints in space within the permitted flight dome. The environment ends when either the Quadcopter collides with the ground or exits the permitted flight dome.
MORE ARE ON THE WAY
Non-Gymnasium examples
If you're not interested in RL but want to use the library for your own research, we provide a bunch of template scripts in examples/ that you can run with python3 examples/***.py in macOS and Linux.
The library is built using CrazyFlie drones, check out the documentation.
These scripts are built with as little dependencies as possible, but enable interfacing with real (using the CrazyPA module) or virtual drones easy.
Simulation Only
sim_single.py
Simulates a single drone in the pybullet env with position control.
sim_swarm.py
Simulates a swarm of drones in the pybullet env with velocity control.
sim_cube.py
Simulates a swarm of drones in a spinning cube.
Hardware Only
fly_single.py
Flies a real Crazyflie, check out the documentation and how to connect to get your URI(s) and modify them in line 18.
fly_swarm.py
Flies a real Crazyflie swarm, same as the previous example, but now takes in a list of URIs.
Simulation or Hardware
sim_n_fly_single.py
Simple script that can be used to fly a single crazyflie in sim or with a real drone using either the --hardware or --simulate args.
sim_n_fly_multiple.py
Simple script that can be used to fly a swarm of crazyflies in sim or with real drones using either the --hardware or --simulate args.
sim_n_fly_cube_from_scratch.py
Simple script that can be used to fly a swarm of crazyflies in sim or with real drones using either the --hardware or --simulate args, and forms the same spinning cube from takeoff as in sim_cube.py.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file PyFlyt-0.3.0.tar.gz.
File metadata
- Download URL: PyFlyt-0.3.0.tar.gz
- Upload date:
- Size: 141.8 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.10.6
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 11e4c08c12fc43fb181db8471011963417b509326a285e062dbc1549d39c30e2 |
|
| MD5 | d0ccc25dea390076731f5b837597ae5e |
|
| BLAKE2b-256 | 3247b24ee53b0ab49444a048e3a8d30c767ba0e76281d93d5cdabe6eee103b5f |
Provenance
File details
Details for the file PyFlyt-0.3.0-py3-none-any.whl.
File metadata
- Download URL: PyFlyt-0.3.0-py3-none-any.whl
- Upload date:
- Size: 157.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.10.6
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | df16c38bebf37eb72262264b5e40ade63c5a84c04edb8073582a4df3768b3d09 |
|
| MD5 | c49518fd76c9012f5a1c8127440095fb |
|
| BLAKE2b-256 | 92502c6e73b846a204941475784c5685663a721bebfcad7c70405e9c60e89323 |
