evaluating-cosmos-policy

# Cosmos Policy Evaluation

Evaluation workflows for NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments from the public `cosmos-policy` repository. Covers blank-machine setup, headless GPU evaluation, and inference profiling.

## Quick start

Run a minimal LIBERO evaluation using the official public eval module:

```bash
uv run --extra cu128 --group libero --python 3.10 \
  python -m cosmos_policy.experiments.robot.libero.run_libero_eval \
    --config cosmos_predict2_2b_480p_libero__inference_only \
    --ckpt_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B \
    --config_file cosmos_policy/config/config.py \
    --use_wrist_image True \
    --use_proprio True \
    --normalize_proprio True \
    --unnormalize_actions True \
    --dataset_stats_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B/libero_dataset_statistics.json \
    --t5_text_embeddings_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B/libero_t5_embeddings.pkl \
    --trained_with_image_aug True \
    --chunk_size 16 \
    --num_open_loop_steps 16 \
    --task_suite_name libero_10 \
    --num_trials_per_task 1 \
    --local_log_dir cosmos_policy/experiments/robot/libero/logs/ \
    --seed 195 \
    --randomize_seed False \
    --deterministic True \
    --run_id_note smoke \
    --ar_future_prediction False \
    --ar_value_prediction False \
    --use_jpeg_compression True \
    --flip_images True \
    --num_denoising_steps_action 5 \
    --num_denoising_steps_future_state 1 \
    --num_denoising_steps_value 1 \
    --data_collection False
```

## Core concepts

**What Cosmos Policy is**: NVIDIA Cosmos Policy is a vision-language-action (VLA) model that uses Cosmos Tokenizer to encode visual observations into discrete tokens, then predicts robot actions conditioned on language instructions and visual context.

**Key architecture choices**:

| Component | Design |
|-----------|--------|
| Visual encoder | Cosmos Tokenizer (discrete tokens) |
| Language conditioning | Cross-attention to language embeddings |
| Action prediction | Autoregressive action token generation |

**Public command surface**: The supported evaluation entrypoints are `cosmos_policy.experiments.robot.libero.run_libero_eval` and `cosmos_policy.experiments.robot.robocasa.run_robocasa_eval`. Keep reproduction notes anchored to these public modules and their documented flags.

## Compute requirements

| Task | GPU | VRAM | Typical wall time |
|------|-----|------|-------------------|
| LIBERO smoke eval (1 trial) | 1x A40/A100 | ~16 GB | 5-10 min |
| LIBERO full eval (50 trials) | 1x A40/A100 | ~16 GB | 2-4 hours |
| RoboCasa single-task (2 trials) | 1x A40/A100 | ~18 GB | 10-15 min |
| RoboCasa all-tasks | 1x A40/A100 | ~18 GB | 4-8 hours |

## When to use vs alternatives

**Use this skill when:**
- Evaluating NVIDIA Cosmos Policy on LIBERO or RoboCasa benchmarks
- Profiling inference latency and throughput for Cosmos Policy
- Setting up headless EGL rendering for robot simulation on GPU clusters

**Use alternatives when:**
- Training or fine-tuning Cosmos Policy from scratch (use official Cosmos training docs)
- Working with OpenVLA-based policies (use `fine-tuning-openvla-oft`)
- Working with Physical Intelligence pi0 models (use `fine-tuning-serving-openpi`)
- Running real-robot evaluation rather than simulation

---

## Workflow 1: LIBERO evaluation

Copy this checklist and track progress:

```text
LIBERO Eval Progress:
- [ ] Step 1: Install environment and dependencies
- [ ] Step 2: Configure headless EGL rendering
- [ ] Step 3: Run smoke evaluation
- [ ] Step 4: Validate outputs and parse results
- [ ] Step 5: Run full benchmark if smoke passes
```

**Step 1: Install environment**

```bash
git clone https://github.com/NVlabs/cosmos-policy.git
cd cosmos-policy
# Follow SETUP.md to build and enter the supported Docker container.
# Then, inside the container:
uv sync --extra cu128 --group libero --python 3.10
```

**Step 2: Configure headless rendering**

```bash
export CUDA_VISIBLE_DEVICES=0
export MUJOCO_EGL_DEVICE_ID=0
export MUJOCO_GL=egl
export PYOPENGL_PLATFORM=egl
```

**Step 3: Run smoke evaluation**

```bash
uv run --extra cu128 --group libero --python 3.10 \
  python -m cosmos_policy.experiments.robot.libero.run_libero_eval \
    --config cosmos_predict2_2b_480p_libero__inference_only \
    --ckpt_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B \
    --config_file cosmos_policy/config/config.py \
    --use_wrist_image True \
    --use_proprio True \
    --normalize_proprio True \
    --unnormalize_actions True \
    --dataset_stats_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B/libero_dataset_statistics.json \
    --t5_text_embeddings_path nvidia/Cosmos-Policy-LIBERO-Predict2-2B/libero_t5_embeddings.pkl \
    --trained_with_image_aug True \
    --chunk_size 16 \
    --num_open_loop_steps 16 \
    --task_suite_name libero_10 \
    --num_trials_per_task 1 \
    --local_log_dir cosmos_policy/experiments/robot/libero/logs/ \
    --seed 195 \
    --randomize_seed False \
    --deterministic True \
    --run_id_note smoke \
    --ar_future_prediction False \
    --ar_value_prediction False \
    --use_jpeg_compression True \
    --flip_images True \
    --num_denoising_steps_action 5 \
    --num_denoising_steps_future_state 1 \
    --num_denoising_steps_value 1 \
    --data_collection False
```

**Step 4: Validate and parse results**

```python
import json
import glob

# Find latest evaluation result from the official log directory
log_files = sorted(glob.glob("cosmos_policy/experiments/robot/libero/logs/**/*.json", recursive=True))
with open(log_files[-1]) as f:
    results = json.load(f)

print(results)
```

**Step 5: Scale up**

Run across all four LIBERO task suites with 50 trials:

```bash
for suite in libero_spatial libero_object libero_goal libero_10; do
  uv run --extra cu128 --group libero --python 3.10 \
    python -m cosmos_policy.experiments.robot.libero.run_libero_eval \
      --config cosmos_predict2_2b_480p_libero__inference_only \