EmBench-Submission

Ideal-Embody Phase-2 Submission

This repository contains the model weights, environment configurations, and evaluation scripts for Team Ideal-Embody's submission to the CVPR 2026 Embodied Challenge (Phase-2).

The submission covers two main tasks: ALFRED and Navigation. Please follow the respective guidelines below to set up the environments and reproduce the evaluation results on the hidden held-out benchmarks.

📂 Repository Structure

. (EmBench-Submission Root)
├── embench.yml                  # Conda environment file for ALFRED
├── embench_nav.yml              # Conda environment file for Navigation
├── Qwen2.5-VL-3B-ALF/           # Model weights / checkpoints for ALFRED
├── Qwen2.5-VL-3B-NAV/           # Model weights / checkpoints for Navigation
├── alfred/                      # Directory for ALFRED task
│   └── alf_base.sh              # Evaluation script for ALFRED base dataset
├── navigation/                  # Directory for Navigation task
│   └── Navgation/               # Core Navigation submission directory
│       └── nav_base.sh          # Evaluation script for Navigation base dataset
└── running/                     # Running logs / outputs

---

🛠️ Task 1: ALFRED Task Instruction

Step 1: Environment Setup

Ensure you are in the repository root directory, then create and activate the dedicated Conda environment:

# Create the environment using the yml file in the root
conda env create -f embench.yml

# Activate the environment
conda activate embench

Step 2: Navigate and Run Evaluation

Change your current working directory to the alfred folder and execute the evaluation script:

# Move to alfred directory
cd alfred

# Run the evaluation on the base dataset
bash alf_base.sh

---

🛠️ Task 2: Navigation Task Instruction

Step 1: Environment Setup

Ensure you are in the repository root directory, then create and activate the dedicated Conda environment:

# Create the environment using the yml file in the root
conda env create -f embench_nav.yml

# Activate the environment
conda activate embench_nav

Step 2: Navigate and Run Evaluation

Change your current working directory to the Navigation submission folder and execute the evaluation script:

# Move to the navigation directory
cd navigation/Navgation/

# Run the evaluation on the base dataset
bash nav_base.sh

---