This repository provides a PyBullet-based OpenAI Gym environment for simulating quadrupedal locomotion, specifically the Spot Mini robot. It targets researchers and developers in reinforcement learning and robotics, enabling the testing and training of gaits without risking physical hardware. The core innovation is D^2 Gait Modulation with Bezier Curves, a method that optimizes existing gait schemes for improved robustness and adaptability.

How It Works

The simulator utilizes PyBullet for physics and rendering, offering a Gym-compatible interface for RL tasks. Gait patterns are generated using 12-point Bezier curves, allowing for smooth and continuous modulation of foot placement and body movements. The D^2 Gait Modulation approach uses a baseline gait and optimizes it through RL, with domain randomization (terrain, mass, friction) applied during training to enhance sim-to-real transfer.

Quick Start & Requirements

• Install/Run: roslaunch mini_ros spot_move.launch (for ROS control) or ./env_tester.py (for non-ROS testing).
• Prerequisites: ROS Melodic, PyTorch, PyBullet, OpenAI Gym, OpenCV, SciPy, NumPy. Joystick recommended for ROS control.
• Links: Docs, Demo

Highlighted Details

• D^2 Gait Modulation with Bezier Curves for RL-based gait optimization.
• Domain randomization for terrain, link masses, and foot friction during training.
• Action space includes clearance height, body height, and foot XYZ residual modulations.
• Supports joystick control via ROS or a GUI-based testing environment.

Maintenance & Community

Development was halted in November 2020 due to an NDA. The project was featured in Robotics Weekly and Mithi's Robotics Coursework.

Licensing & Compatibility

The repository does not explicitly state a license. The citation lists the year 2020 and version 2.1.0.

Limitations & Caveats

Development was halted in November 2020. The ROS Melodic dependency may be outdated. The project's primary focus is on gait modulation, and other aspects of robot control might be less developed.