This project provides open-source robot simulation environments integrated with OpenAI Gym, aimed at researchers and developers in reinforcement learning and robotics. It offers a suite of physics-based tasks for training and evaluating robotic agents, with a focus on replicating and extending existing benchmarks.

How It Works

Roboschool utilizes a custom-built Bullet Physics engine and a C++ rendering engine for efficient simulation. It exposes environments through the OpenAI Gym API, allowing seamless integration with standard RL algorithms. The simulation focuses on realistic physics and complex locomotion tasks, moving beyond simpler trajectory-centric problems.

Quick Start & Requirements

• Install via pip: pip install roboschool
• For headless machines: sudo apt-get install libgl1-mesa-dev
• Installation from source requires Qt5, Boost-Python3, Assimp, CMake, and patchelf.
• Detailed installation instructions and helper scripts (install_boost.sh, install_bullet.sh, exports.sh) are provided for Linux and macOS.
• Official blog post: https://blog.openai.com/roboschool/

Highlighted Details

• Includes 11 distinct Roboschool environments (e.g., RoboschoolHumanoidFlagrun-v1, RoboschoolAnt-v1).
• Offers pre-trained agents in the agent_zoo directory for demonstration.
• Supports keyboard-controlled humanoid simulation.
• Version 1.1 introduced stronger joint punishment and improved learning schedules for specific environments.

Maintenance & Community

• Status: DEPRECATED and Archived. The project is provided as-is with no further updates expected.
• OpenAI recommends using PyBullet instead.

Licensing & Compatibility

• The project appears to be licensed under the MIT License.
• Compatible with Python 3.5 and 3.6 on OS X and Linux. Python 2.7 support may require significant effort.

Limitations & Caveats

This project is officially deprecated and archived, meaning no further development or support is provided. Users are strongly advised to migrate to PyBullet for ongoing projects. Installation from source can be complex due to numerous system-level dependencies.