A Minimal Cognitive Architecture

Features

• BriCA
• OpenAI Gym
• Perceptual Module + Motor Module
• Autoencoder for Perceptual Module
• PPO for Motor Module
• Curriculum Learning

For the details, please read this article.

How to Install

• Clone the repository
• BriCA1
  • Follow the instruction here.
• Cerenaut Core
  • Follow the instruction here.
• pip install the following
  • OpenAI Gym
    • pip install gym
  • TensorForce
    • pip install tensorforce
  • TensorFlow (used in TensorForce and TensorBoard)
    • pip install tensorflow

To train a perceptual model

Dumping perceptual data

$ python minimal_CA.py 2 --config minimal_CA.json --episode_count 250 --max_steps 1500 --dump dump.txt

This feeds the Gym environment observation directly to the reinforcement learnig to cause learning, while dumping the observation to the dump file.

Training a perceptual model

$ python gym_test_simple.py --dataset dump.txt --save-model saved_model.pt --config minimal_CA.json

This is to train the simple autoencoder from cerenaut_core and to output the model file.

Reinforcement learning with the trained perceptual model

$ python minimal_CA.py 2 --config minimal_CA.json --episode_count 50 --max_steps 1500 --model saved_model.pt