The Balloon Learning Environment - flying stratospheric balloons with deep reinforcement learning.

Last update: Dec 25, 2022

Overview

Balloon Learning Environment

The Balloon Learning Environment (BLE) is a simulator for stratospheric balloons. It is designed as a benchmark environment for deep reinforcement learning algorithms, and is a followup to the Nature paper "Autonomous navigation of stratospheric balloons using reinforcement learning".

Getting Started

Note: The BLE requires python >= 3.7

The BLE can easily be installed with pip:

pip install --upgrade pip && pip install balloon_learning_environment

Once the package has been installed, you can test it runs correctly by evaluating one of the benchmark agents:

python -m balloon_learning_environment.eval.eval \
  --agent=station_seeker \
  --renderer=matplotlib \
  --suite=micro_eval \
  --output_dir=/tmp/ble/eval

Ensure the BLE is Using Your GPU/TPU

The BLE contains a VAE for generating winds, which you will probably want to run on your accelerator. See the jax documentation for installing with GPU or TPU.

As a sanity check, you can open interactive python and run:

from balloon_learning_environment.env import balloon_env
env = balloon_env.BalloonEnv()

If you are not running with GPU/TPU, you should see a log like:

WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)

If you don't see this log, you should be good to go!

Next Steps

For more information, see the docs.

Giving credit

If you use the Balloon Learning Environment in your work, we ask that you use the following BibTeX entry:

@software{Greaves_Balloon_Learning_Environment_2021,
  author = {Greaves, Joshua and Candido, Salvatore and Dumoulin, Vincent and Goroshin, Ross and Ponda, Sameera S. and Bellemare, Marc G. and Castro, Pablo Samuel},
  month = {12},
  title = {{Balloon Learning Environment}},
  url = {https://github.com/google/balloon-learning-environment},
  version = {1.0.0},
  year = {2021}
}

If you use the ble_wind_field dataset, you should also cite

Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A.,
Muñoz‐Sabater, J., Nicolas, J., Peubey, C., Radu, R., Schepers, D., Simmons, A.,
Soci, C., Abdalla, S., Abellan, X., Balsamo, G., Bechtold, P., Biavati, G.,
Bidlot, J., Bonavita, M., De Chiara, G., Dahlgren, P., Dee, D., Diamantakis, M.,
Dragani, R., Flemming, J., Forbes, R., Fuentes, M., Geer, A., Haimberger, L.,
Healy, S., Hogan, R.J., Hólm, E., Janisková, M., Keeley, S., Laloyaux, P.,
Lopez, P., Lupu, C., Radnoti, G., de Rosnay, P., Rozum, I., Vamborg, F.,
Villaume, S., Thépaut, J-N. (2017): Complete ERA5: Fifth generation of ECMWF
atmospheric reanalyses of the global climate. Copernicus Climate Change Service
(C3S) Data Store (CDS). (Accessed on 01-04-2021)

The Balloon Learning Environment - flying stratospheric balloons with deep reinforcement learning.

Related tags

Overview

Balloon Learning Environment

Getting Started

Ensure the BLE is Using Your GPU/TPU

Next Steps

Giving credit

Owner

Google

KinectFusion implemented in Python with PyTorch

Collection of tasks for fast prototyping, baselining, finetuning and solving problems with deep learning.

PyTorch implementation of "LayoutTransformer: Layout Generation and Completion with Self-attention"

High performance distributed framework for training deep learning recommendation models based on PyTorch.

Official implementation of particle-based models (GNS and DPI-Net) on the Physion dataset.

TorchMD-Net provides state-of-the-art graph neural networks and equivariant transformer neural networks potentials for learning molecular potentials

Instance-based label smoothing for improving deep neural networks generalization and calibration

A collection of 100 Deep Learning images and visualizations

Statistical-Rethinking-with-Python-and-PyMC3 - Python/PyMC3 port of the examples in " Statistical Rethinking A Bayesian Course with Examples in R and Stan" by Richard McElreath

Generative Exploration and Exploitation - This is an improved version of GENE.

Fader Networks: Manipulating Images by Sliding Attributes - NIPS 2017

UNAVOIDS: Unsupervised and Nonparametric Approach for Visualizing Outliers and Invariant Detection Scoring

Implementation for the paper: Invertible Denoising Network: A Light Solution for Real Noise Removal (CVPR2021).

simple demo codes for Learning to Teach with Dynamic Loss Functions

Sarus implementation of classical ML models. The models are implemented using the Keras API of tensorflow 2. Vizualization are implemented and can be seen in tensorboard.

EMNLP'2021: Simple Entity-centric Questions Challenge Dense Retrievers

RANZCR-CLiP 7th Place Solution

WarpRNNT loss ported in Numba CPU/CUDA for Pytorch

KIDA: Knowledge Inheritance in Data Aggregation

Sound Source Localization for AI Grand Challenge 2021