Deep Skipper discovers minimal fuel cost trajectories from Origin to Destination. In this repository we showcase minimal fuel cost path from Panama to Osaka. The technique used for exploration is based on Deep Reinforcement Learning with Policy Gradient based learning.
In order to facilitate this experimentation, we created an OpenAI Gym environment called, Shipping. We also created a common library called nautlabs. We encourage you to write your own agent code for this environment. Our simple policy gradient based implementation is simple_pg.py. An evaluation code that corresponds to that trained model is in eval_simple.py.
The data that is required for this setup is the Global Marine weather data from National Centers for Environmental Information:
- wind.json (also known as GFS)
- wave-height.json (also known as WAVE-HEIGHT)
- oscar.json (ocean current information)
Following Python packages are required for this suite of the programs to run:
pip==10.0.1
numpy==1.15.4
pandas==0.23.0
matplotlib==3.0.3
jsonschema==2.6.0
basemap==1.2.0
basemap-data-hires==1.2.0
boto==2.48.0
botocore==1.12.169
boto3==1.9.146
awscli==1.16.179
geographiclib==1.49
tensorflow==1.13.1
tensorflow-estimator==1.13.0
tensorflow-serving-api==1.13.0
Keras==2.2.4
gym==0.12.5
ffmpeg==4.0
pip install -e gym_shipping
Execute file runp.sh:
.
├── index.md
├── map_animate.ipynb
├── nav2.mp4
├── requirements.txt
├── runp.sh
The contents here include:
.
├── gym-shipping
│ ├── README.md
│ ├── __init__.py
│ ├── gym_shipping
│ │ ├── __init__.py
│ │ ├── envs
│ │ │ ├── __init__.py
│ │ │ ├── shipping_env.py
│ │ └── test
│ │ ├── __init__.py
│ │ └── test_shipping_env.py
│ ├── gym_shipping.egg-info
│ │ ├── PKG-INFO
│ │ ├── SOURCES.txt
│ │ ├── dependency_links.txt
│ │ ├── requires.txt
│ │ └── top_level.txt
│ └── setup.py
.
├── nautlabs
│ ├── __init__.py
│ ├── shipperf.py
│ └── tests
│ └── test_shipperf.py
.
├── eval_simple.py
└── simple_pg.py
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent