Cosim

2.1 Overview

cosim is a Sim-to-Sim Framework built on the MuJoCo physics engine. It is designed to study transitions between different simulation environments and evaluate model performance under varying physical and sensory conditions.

^{2.2 Key Features}

• 📌 MuJoCo-Powered Simulation High-performance, physics-based simulation for precise and realistic modeling.

• 🔁 Cross-Environment Transfer Supports transfer learning and benchmarking across different physical configurations and terrains.

• 📈 Comprehensive Analysis Tools Automatically generates detailed reports for policy evaluation and performance analysis.

• ⚙️ Configurable Physical Properties

  • Sliding, rolling, and torsional friction

  • Mass noise, load variance

  • Action delay probability

  • Actuator gains (Kp/Kd for hip, shoulder, leg, wheel)

  • ... and more!

• 🌐 Diverse Terrain Support Includes predefined environments:

  • flat, rocky_easy, rocky_hard

  • slope_easy, slope_hard

  • stairs_up_easy, stairs_up_normal, stairs_up_hard

• 🎮 Dual Command Modes

  • Velocity-based commands

  • Position-based commands

• 🗺️ Height Map Integration Real-time terrain visualization with adjustable map size and resolution.

• 🧩 Modular User Interface

  • Environment & policy configuration

  • Hardware tuning

  • Key-based command input

  • Sensor noise and randomization settings

^{2.3 Installation}

• 1. Create a Conda Environment (Python 3.10 recommended)

conda create -n cosim python=3.10
conda activate cosim

• 2. Install Dependencies

pip install -r requirements.txt

• 3. Run the Simulator

python launch.py

^{2.4 Project Structure}

cosim/
├── core/                 # Policy interface and testing logic
├── envs/                 # Environment wrapper and simulation backend
├── config/               # YAML configuration files
├── docs/                 # docs
├── gui/                  # Main PyQt5-based user interface
├── launch.py             # Entry point to run GUI
└── README.md