HoMeR

---

Overview

HoMeR (Hybrid Whole-Body Policies for Mobile Robots) is a hybrid imitation learning framework for mobile manipulation. It combines whole-body control with a hybrid action representation to achieve generalizable and precise robot behavior in both simulation and real-world settings.

---

Quick Start

Depending on your use case, please follow the appropriate setup and usage instructions:

🖥️ Simulation-Only

If you are only using simulation, refer to:

📄 SIM.md

This guide covers:

• Conda setup on macOS and Linux
• Simulated data collection and annotation
• Training and evaluating HoMeR and baselines in simulation

---

🤖 Real-World

If you plan to use HoMeR in real, refer to:

📄 REAL.md

This guide covers:

• Hardware and software setup for real-world deployment
• Real-world data collection and annotation
• Training and evaluating HoMeR and baselines in real

---

Repository Structure

cfgs/                 # Training config files
envs/                 # Environment setup for sim and real
docker/               # Real-world Docker setup
scripts/              # Training and evaluation scripts
interactive_scripts/  # Data collection, replay, and data annotation tools
dataset_utils/        # Dataset loading and data visualization tools
mj_assets/            # MJCF assets for simulation
sbatch_scripts/       # SLURM scripts to launch training jobs