inverse_rl  by justinjfu

Inverse RL implementations for imitation learning algorithms

created 8 years ago
274 stars

Top 95.2% on sourcepulse

GitHubView on GitHub
Project Summary

This repository provides implementations of Inverse Reinforcement Learning (IRL) and imitation learning algorithms, specifically GAIL and Guided Cost Learning (GCL), for researchers and practitioners in reinforcement learning. It aims to enable learning cost functions from expert demonstrations to reproduce desired behaviors.

How It Works

The library implements algorithms like Generative Adversarial Imitation Learning (GAIL) and Guided Cost Learning (GCL), which leverage deep learning frameworks to learn policies from expert trajectories. GAIL uses a discriminator to distinguish between expert and generated trajectories, while GCL learns a cost function that explains the expert's behavior.

Quick Start & Requirements

  • Install: Requires rllab and tensorflow.
  • Prerequisites: rllab (https://github.com/openai/rllab), tensorflow.
  • Example: Run scripts/pendulum_data_collect.py to collect expert data for Pendulum-v0, then scripts/pendulum_gcl.py to run GCL. Expected average return for Pendulum-v0 is around -100 to -150.

Highlighted Details

  • Implements GAIL (Generative Adversarial Imitation Learning).
  • Implements Guided Cost Learning (GCL) with GAN formulation.
  • Includes Tabular MaxCausalEnt IRL.

Maintenance & Community

No information on maintenance or community channels is provided in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility with commercial or closed-source projects is unknown.

Limitations & Caveats

The project depends on rllab, which is an older framework and may have compatibility issues with current deep learning libraries or Python versions. The README does not mention specific version requirements for dependencies.

Health Check
Last commit

7 years ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
3 stars in the last 90 days

Explore Similar Projects

Feedback? Help us improve.