Skip to content

Tensorflow implementation of DeepMind paper - "Learning to Navigate in Complex Environments"

Notifications You must be signed in to change notification settings

tgangwani/GA3C-DeepNavigation

 
 

Repository files navigation

Maze Navigation using Reinforcement Learning

Description

TensorFlow implementation of ideas in the DeepMind paper "Learning to Navigate in Complex Environments". The baseline architecture is GPU-based A3C from the paper "Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU". Full report.

Requirements

  • TensorFlow 1.0
  • DeepMind Lab
  • Python plugins - numpy, cv2, pygame

Getting Started

Dowload and install DeepMind Lab

$ git clone https://github.com/deepmind/lab.git

Build it following the build instructions

Clone repo inside the lab directory

$ cd lab
$ git clone https://github.com/tgangwani/GA3C-DeepNavigation.git

Add the bazel instructions at the end of lab/BUILD file

py_binary(
    name = "GA3C-DeepNavigation_train",
    srcs = ["GA3C-DeepNavigation/GA3C.py"],
    data = [":deepmind_lab.so"],
    main = "GA3C-DeepNavigation/GA3C.py",
)

Then run bazel command to run the agent

bazel run :GA3C-DeepNavigation_train --define headless=osmesa

Use PLAY_MODE=False in Config.py for training. Setting PLAY_MODE=True loads model parameters from a checkpoint and runs a single agent. A display is expected.

Network

The neural net architecture is the same as in the paper, but for the loop-prediction loss.

Results

1. Nav_maze_static_01

Live Agent

YouTube

Learning Curve

2. Stairway to melon

Live Agent

YouTube

Learning Curve

Acknowledgement

Unreal code by miyosuda

About

Tensorflow implementation of DeepMind paper - "Learning to Navigate in Complex Environments"

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%