Additional environments compatible with OpenAI gym

Last update: Dec 06, 2022

Related tags

Deep Learning quad-swarm-rl

Overview

Decentralized Control of Quadrotor Swarms with End-to-end Deep Reinforcement Learning

A codebase for training reinforcement learning policies for quadrotor swarms. Includes:

Flight dynamics simulator forked from https://github.com/amolchanov86/gym_art and extended to support swarms of quadrotor drones
Scripts and the necessary wrappers to facilitate training of control policies with Sample Factory https://github.com/alex-petrenko/sample-factory

Paper: https://arxiv.org/abs/2109.07735

Website: https://sites.google.com/view/swarm-rl

Installation

Initialize a Python environment, i.e. with conda (Python versions 3.6-3.8 are supported):

conda create -n swarm-rl python=3.8
conda activate swarm-rl

Clone and install this repo as an editable Pip package:

git clone https://github.com/alex-petrenko/quad-swarm-rl
cd quad-swarm-rl
pip install -e .

This should pull and install all the necessary dependencies, including Sample Factory and PyTorch.

Running experiments

Train

This will run the baseline experiment. Change the number of workers appropriately to match the number of logical CPU cores on your machine, but it is advised that the total number of simulated environments is close to that in the original command:

python -m swarm_rl.train --env=quadrotor_multi --train_for_env_steps=1000000000 --algo=APPO \
--use_rnn=False \
--num_workers=36 --num_envs_per_worker=4 \
--learning_rate=0.0001 --ppo_clip_value=5.0 \
--recurrence=1 --nonlinearity=tanh --actor_critic_share_weights=False \
--policy_initialization=xavier_uniform --adaptive_stddev=False --with_vtrace=False \
--max_policy_lag=100000000 --hidden_size=256 --gae_lambda=1.00 --max_grad_norm=5.0 \
--exploration_loss_coeff=0.0 --rollout=128 --batch_size=1024 --quads_use_numba=True \
--quads_mode=mix --quads_episode_duration=15.0 --quads_formation_size=0.0 \
--encoder_custom=quad_multi_encoder --with_pbt=False --quads_collision_reward=5.0 \
--quads_neighbor_hidden_size=256 --neighbor_obs_type=pos_vel --quads_settle_reward=0.0 \
--quads_collision_hitbox_radius=2.0 --quads_collision_falloff_radius=4.0 --quads_local_obs=6 \
--quads_local_metric=dist --quads_local_coeff=1.0 --quads_num_agents=8 --quads_collision_reward=5.0 \
--quads_collision_smooth_max_penalty=10.0 --quads_neighbor_encoder_type=attention \
--replay_buffer_sample_prob=0.75 --anneal_collision_steps=300000000 --experiment=swarm_rl

Or, even better, you can use the runner scripts in swarm_rl/runs/. Runner scripts (a Sample Factory feature) are Python files that contain experiment parameters, and support features such as evaluation on multiple seeds and gridsearches.

To execute a runner script run the following command:

python -m sample_factory.runner.run --run=swarm_rl.runs.quad_multi_mix_baseline_attn --runner=processes --max_parallel=4 --pause_between=1 --experiments_per_gpu=1 --num_gpus=4

This command will start training four different seeds in parallel on a 4-GPU server. Adjust the parameters accordingly to match your hardware setup.

To monitor the experiments, go to the experiment folder, and run the following command:

tensorboard --logdir=./

Test

To test the trained model, run the following command:

python -m swarm_rl.enjoy --algo=APPO --env=quadrotor_multi --replay_buffer_sample_prob=0 --continuous_actions_sample=False --quads_use_numba=False --train_dir=PATH_TO_PROJECT/swarm_rl/train_dir --experiments_root=EXPERIMENT_ROOT --experiment=EXPERIMENT_NAME

Unit Tests

To run unit tests:

./run_tests.sh

Additional environments compatible with OpenAI gym

Related tags

Overview

Decentralized Control of Quadrotor Swarms with End-to-end Deep Reinforcement Learning

Installation

Running experiments

Train

Test

Unit Tests

Owner

Zhehui Huang

Validated, scalable, community developed variant calling, RNA-seq and small RNA analysis

Safe Model-Based Reinforcement Learning using Robust Control Barrier Functions

A minimalist implementation of score-based diffusion model

Learning High-Speed Flight in the Wild

Radar-to-Lidar: Heterogeneous Place Recognition via Joint Learning

An end-to-end machine learning library to directly optimize AUC loss

Official implementation of the paper "Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering"

A generator of point clouds dataset for PyPipes.

Learning to Draw: Emergent Communication through Sketching

Ascend your Jupyter Notebook usage

Classification models 1D Zoo - Keras and TF.Keras

ObsPy: A Python Toolbox for seismology/seismological observatories.

GT China coal model

A Comprehensive Study on Learning-Based PE Malware Family Classification Methods

MlTr: Multi-label Classification with Transformer

Books, Presentations, Workshops, Notebook Labs, and Model Zoo for Software Engineers and Data Scientists wanting to learn the TF.Keras Machine Learning framework

Code for Iso-Points: Optimizing Neural Implicit Surfaces with Hybrid Representations

This is a virtual picture dragging application. Users may virtually slide photos across the screen. The distance between the index and middle fingers determines the movement. Smaller distances indicate click and motion, whereas bigger distances indicate only hand movement.

Official implementation of the MM'21 paper Constrained Graphic Layout Generation via Latent Optimization

Patch-Based Deep Autoencoder for Point Cloud Geometry Compression