Accelerating Human-AI Co-adaptation (Multi-agent Group 12)

This repository is for Accelerating Human-AI Co-adaptation. Adapted our code from 'https://github.com/VT-Collab/RILI_co-adaptation/tree/main' which is the main repository for the paper "Learning Latent Representations to Co-Adapt to Humans" [http://arxiv.org/abs/2212.09586].

In this repository we include codes for the two strategies:

• RILI vs SAC
• Multiple Agents

Instructions

You can install the packages by running the following command

pip install -r requirements.txt

Then, install the gym environment from any folder:

cd multiple_agents
cd 2Agents
cd gym-rili
pip install -e .
cd ..

Codes for RILI vs SAC

• main.py : Contains code for training the Hider and Seeker in Circle environment
• replaymemory.py : Contains code for storing RILI agent's memory
• replaymemory_SAC.py : Contains code for storing memory of SAC agent
• models/: directory for saved agents
• runs/: directory to visualize losses and rewards using tensorboard --logdir runs
• gym-rili/gym_rili/envs/circle.py: main environment for our code
• algos/sac_agent.py and algos/sac_model_networks.py: Codes for the SAC Agent
• algos/rili.py, algos/model_rili.py and algos/model_sac.py: Codes for the RILI Agent

Code files for Multiple Agents

common subfolders

algos

• code for RILI Agent

2_Agents

• main.py : Contains code for pretraining the model with 2 agents
• maintest.py : Contains code for testing the pre-trained model on circle environment
• replaymemory.py : Contains code for storing memory
• env/circle.py : Contains code for circle-N environment used to pretrain 3 agents

3_Agents

• main.py : Contains code for pretraining the model with 3 agents
• maintest.py : Contains code for testing the pre-trained model on circle environment
• replaymemory.py : Contains code for storing memory
• env/circle.py : Contains code for circle-N environment used to pretrain 3 agents