Skip to content
BranchesTags
 
 

Repository files navigation

Instructions for training on LIBERO:

First follow the repo install instructions. Setup UV in a virtual environment in this repo before uv installing things. Make sure conda is deactivated!!!!!

Basic Install:

uv venv
source .venv/bin/activate
GIT_LFS_SKIP_SMUDGE=1 uv sync
uv pip install tensorflow tensorflow_datasets shapely openai # openai is for the maskedvla evaluation
uv pip install git+https://github.com/memmelma/vila_utils.git

Modified LIBERO Dataset:

Follow the instructions in my openvla repo to install and generate the modified LIBERO dataset: here.

There are two ways to convert the LIBERO dataset to a LeRobot dataset:

Option 1: Basic Conversion

Convert the basic modified LIBERO dataset to a LeRobot dataset:

uv run examples/libero/convert_libero_data_to_lerobot.py --data_dir /home/jeszhang/tensorflow_datasets/

Option 2: Conversion with Path Masks and Subtasks

For more advanced training with path masks and subtask instructions, you can use the following script. note that the data_dir expects the data dir of the OpenVLA processed hdf5 LIBERO data, not the tensorflow dataset. Download from here.

uv run examples/libero/convert_pathmask_libero_data_to_lerobot.py --data_dir /home1/jessez/scratch_data/libero_openvla_processed_datasets --path_and_mask_file_dir /home1/jessez/project2_data/libero_90_processed_256

This script supports several additional options:

  • --use_subtask_instructions: Divides episodes by subtask instructions instead of using full task instructions
  • --push_to_hub: Pushes the processed dataset to Hugging Face Hub
  • --return_full_path_mask: Uses full trajectory paths and masks instead of the subtask path masks

The path mask conversion relies on annotation files containing masks, paths, and subtask information.

I've already processed the training stats and norms in assets/pi0_libero_low_mem_finetune/jesbu1/libero_90_lerobot/norm_stats.json. If not running pi0_libero_low_mem_finetune, you can copy the norm_stats.json file to the assets/CONFIG_NAME/jesbu1/libero_90_lerobot/ directory.

Training π₀ with LoRA

You may need to change the repo_id in the src/openpi/training/config.py file for the pi0_libero_low_mem_finetune config to the jesbu1/libero_90_lerobot dataset and where it is on your machine. It should by default be in /home/$USER/.cache/huggingface/lerobot/jesbu1/libero_90_lerobot/. You can also change the local_files_only flag to False in the src/openpi/training/config.py file to use the local dataset.

Basic Training

Then train:

XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 uv run scripts/train.py pi0_libero_low_mem_finetune --exp-name=EXP_NAME --overwrite

Training with Validation

To train with validation, you can specify a validation dataset in the config. The validation dataset should be a separate LeRobot dataset that follows the same format as your training data. The validation will run every validation_interval steps (default 1000) and log the validation metrics to wandb.

Example config modification:

config = TrainConfig(
    # ... other config ...
    validation_data=LeRobotLiberoDataConfig(
        repo_id="jesbu1/libero_90_lerobot_val",  # Your validation dataset
        base_config=DataConfig(
            local_files_only=True,
            prompt_from_task=True,
        ),
    ),
    validation_interval=1000,  # Run validation every 1000 steps
)

Training with Path Masks

You can also train with path masks by running the following:

XLA_PYTHON_CLIENT_MEM_FRACTION=0.95 uv run scripts/train.py pi0_libero_low_mem_finetune_path --exp-name=EXP_NAME --overwrite
XLA_PYTHON_CLIENT_MEM_FRACTION=0.95 uv run scripts/train.py pi0_libero_low_mem_finetune_masked --exp-name=EXP_NAME --overwrite
XLA_PYTHON_CLIENT_MEM_FRACTION=0.95 uv run scripts/train.py pi0_libero_low_mem_finetune_path_masked --exp-name=EXP_NAME --overwrite

XLA_PYTHON_CLIENT_MEM_FRACTION=0.95  uv run scripts/train.py pi0_libero_low_mem_finetune_path_no_proprio --exp-name=EXP_NAME --overwrite
XLA_PYTHON_CLIENT_MEM_FRACTION=0.95  uv run scripts/train.py pi0_libero_low_mem_finetune_masked_no_proprio --exp-name=EXP_NAME --overwrite
XLA_PYTHON_CLIENT_MEM_FRACTION=0.95  uv run scripts/train.py pi0_libero_low_mem_finetune_path_masked_no_proprio --exp-name=EXP_NAME --overwrite

Evaluation of LIBERO

Once done training, you can evaluate the model by running the following command to initialize a policy server:

CUDA_VISIBLE_DEVICES=1 uv run scripts/serve_policy.py policy:checkpoint --policy.config=pi0_libero_low_mem_finetune --policy.dir=checkpoints/pi0_libero_90_LoRA_finetune_8gpu/29999/ 

uv run scripts/serve_policy.py policy:checkpoint --policy.config=pi0_libero_low_mem_finetune_path --policy.dir=checkpoints/pi0_libero_low_mem_finetune_path/pi0_libero_90_path_bs164_rdp/35000/

uv run scripts/serve_policy.py policy:checkpoint --policy.config=pi0_libero_low_mem_finetune_path_no_proprio --policy.dir=checkpoints/pi0_libero_low_mem_finetune_path_no_proprio/pi0_libero_90_path_no_proprio_bs128/27000

uv run scripts/serve_policy.py policy:checkpoint --policy.config=pi0_libero_low_mem_finetune_masked_no_proprio --policy.dir=checkpoints/pi0_libero_low_mem_finetune_masked_no_proprio/pi0_libero_90_masked_no_proprio_bs128/28000

In a separate terminal, run the following command to run the Libero evaluation script:

# Create virtual environment
conda deactivate
uv venv --python 3.8 examples/libero/.venv
source examples/libero/.venv/bin/activate
uv pip sync examples/libero/requirements.txt third_party/libero/requirements.txt --extra-index-url https://download.pytorch.org/whl/cu113 --index-strategy=unsafe-best-match
uv pip install -e packages/openpi-client
uv pip install -e third_party/libero
uv pip install -e ../vila_utils # from https://github.com/memmelma/vila_utils
uv pip install wandb
uv pip install openai shapely # for pathmask
export PYTHONPATH=$PYTHONPATH:$PWD/third_party/libero

# Run the simulation
python examples/libero/main.py --args.task_suite_name=libero_10 --args.draw_path --args.draw_mask --args.vlm_server_ip="https://whippet-pet-singularly.ngrok.app" --args.vlm_query_frequency=20
python examples/libero/main.py --args.task_suite_name=libero_spatial --args.draw_path --args.draw_mask --args.vlm_server_ip="https://whippet-pet-singularly.ngrok.app" --args.vlm_query_frequency=20
python examples/libero/main.py --args.task_suite_name=libero_object --args.draw_path --args.draw_mask --args.vlm_server_ip="https://whippet-pet-singularly.ngrok.app" --args.vlm_query_frequency=20
python examples/libero/main.py --args.task_suite_name=libero_goal --args.draw_path --args.draw_mask --args.vlm_server_ip="https://whippet-pet-singularly.ngrok.app" --args.vlm_query_frequency=20

python examples/libero/main.py --args.task_suite_name=libero_spatial --args.draw_path --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio"
python examples/libero/main.py --args.task_suite_name=libero_object --args.draw_path --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio"
python examples/libero/main.py --args.task_suite_name=libero_goal --args.draw_path --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio"
python examples/libero/main.py --args.task_suite_name=libero_10 --args.draw_path --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio"


python examples/libero/main.py --args.task_suite_name=libero_spatial --args.draw_path --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8003
python examples/libero/main.py --args.task_suite_name=libero_object --args.draw_path --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8003
python examples/libero/main.py --args.task_suite_name=libero_goal --args.draw_path --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8003
python examples/libero/main.py --args.task_suite_name=libero_10 --args.draw_path --args.vlm_server_ip="http://0.0.0.0:8002" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8003

python examples/libero/main.py --args.task_suite_name=libero_spatial --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8004" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8005
python examples/libero/main.py --args.task_suite_name=libero_object --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8004" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8005
python examples/libero/main.py --args.task_suite_name=libero_goal --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8004" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8005
python examples/libero/main.py --args.task_suite_name=libero_10 --args.draw_mask --args.vlm_server_ip="http://0.0.0.0:8004" --args.vlm_query_frequency=20 --args.wandb_name_suffix="no_proprio" --args.port 8005

openpi

openpi holds open-source models and packages for robotics, published by the Physical Intelligence team.

Currently, this repo contains two types of models:

  • the π₀ model, a flow-based diffusion vision-language-action model (VLA)
  • the π₀-FAST model, an autoregressive VLA, based on the FAST action tokenizer.

For both models, we provide base model checkpoints, pre-trained on 10k+ hours of robot data, and examples for using them out of the box or fine-tuning them to your own datasets.

This is an experiment: $\pi_0$ was developed for our own robots, which differ from the widely used platforms such as ALOHA and DROID, and though we are optimistic that researchers and practitioners will be able to run creative new experiments adapting $\pi_0$ to their own platforms, we do not expect every such attempt to be successful. All this is to say: $\pi_0$ may or may not work for you, but you are welcome to try it and see!

Requirements

To run the models in this repository, you will need an NVIDIA GPU with at least the following specifications. These estimations assume a single GPU, but you can also use multiple GPUs with model parallelism to reduce per-GPU memory requirements by configuring fsdp_devices in the training config. Please also note that the current training script does not yet support multi-node training.

Mode Memory Required Example GPU
Inference > 8 GB RTX 4090
Fine-Tuning (LoRA) > 22.5 GB RTX 4090
Fine-Tuning (Full) > 70 GB A100 (80GB) / H100

The repo has been tested with Ubuntu 22.04, we do not currently support other operating systems.

Installation

When cloning this repo, make sure to update submodules:

git clone --recurse-submodules git@github.com:Physical-Intelligence/openpi.git

# Or if you already cloned the repo:
git submodule update --init --recursive

We use uv to manage Python dependencies. See the uv installation instructions to set it up. Once uv is installed, run the following to set up the environment:

GIT_LFS_SKIP_SMUDGE=1 uv sync

NOTE: GIT_LFS_SKIP_SMUDGE=1 is needed to pull LeRobot as a dependency.

Docker: As an alternative to uv installation, we provide instructions for installing openpi using Docker. If you encounter issues with your system setup, consider using Docker to simplify installation. See Docker Setup for more details.

Model Checkpoints

Base Models

We provide multiple base VLA model checkpoints. These checkpoints have been pre-trained on 10k+ hours of robot data, and can be used for fine-tuning.

Model Use Case Description Checkpoint Path
$\pi_0$ Fine-Tuning Base diffusion π₀ model for fine-tuning s3://openpi-assets/checkpoints/pi0_base
$\pi_0$-FAST Fine-Tuning Base autoregressive π₀-FAST model for fine-tuning s3://openpi-assets/checkpoints/pi0_fast_base

Fine-Tuned Models

We also provide "expert" checkpoints for various robot platforms and tasks. These models are fine-tuned from the base models above and intended to run directly on the target robot. These may or may not work on your particular robot. Since these checkpoints were fine-tuned on relatively small datasets collected with more widely available robots, such as ALOHA and the DROID Franka setup, they might not generalize to your particular setup, though we found some of these, especially the DROID checkpoint, to generalize quite broadly in practice.

Model Use Case Description Checkpoint Path
$\pi_0$-FAST-DROID Inference $\pi_0$-FAST model fine-tuned on the DROID dataset, can perform a wide range of simple table-top manipulation tasks 0-shot in new scenes on the DROID robot platform s3://openpi-assets/checkpoints/pi0_fast_droid
$\pi_0$-DROID Fine-Tuning $\pi_0$ model fine-tuned on the DROID dataset, faster inference than $\pi_0$-FAST-DROID, but may not follow language commands as well s3://openpi-assets/checkpoints/pi0_droid
$\pi_0$-ALOHA-towel Inference $\pi_0$ model fine-tuned on internal ALOHA data, can fold diverse towels 0-shot on ALOHA robot platforms s3://openpi-assets/checkpoints/pi0_aloha_towel
$\pi_0$-ALOHA-tupperware Inference $\pi_0$ model fine-tuned on internal ALOHA data, can unpack food from a tupperware container s3://openpi-assets/checkpoints/pi0_aloha_tupperware
$\pi_0$-ALOHA-pen-uncap Inference $\pi_0$ model fine-tuned on public ALOHA data, can uncap a pen s3://openpi-assets/checkpoints/pi0_aloha_pen_uncap

By default, checkpoints are automatically downloaded from s3://openpi-assets and are cached in ~/.cache/openpi when needed. You can overwrite the download path by setting the OPENPI_DATA_HOME environment variable.

Running Inference for a Pre-Trained Model

Our pre-trained model checkpoints can be run with a few lines of code (here our $\pi_0$-FAST-DROID model):

from openpi.training import config
from openpi.policies import policy_config
from openpi.shared import download

config = config.get_config("pi0_fast_droid")
checkpoint_dir = download.maybe_download("s3://openpi-assets/checkpoints/pi0_fast_droid")

# Create a trained policy.
policy = policy_config.create_trained_policy(config, checkpoint_dir)

# Run inference on a dummy example.
example = {
    "observation/exterior_image_1_left": ...,
    "observation/wrist_image_left": ...,
    ...
    "prompt": "pick up the fork"
}
action_chunk = policy.infer(example)["actions"]

You can also test this out in the example notebook.

We provide detailed step-by-step examples for converting data, training, and running inference on various robots:

Remote Inference: We provide examples and code for running inference of our models remotely: the model can run on a different server and stream actions to the robot via a websocket connection. This makes it easy to use more powerful GPUs off-robot and keep robot and policy environments separate.

Test inference without a robot: We provide a script for testing inference without a robot. This script will generate a random observation and run inference with the model. See here for more details.

Fine-Tuning Base Models on Your Own Data

We will fine-tune the $\pi_0$-FAST model on the Libero dataset as a running example for how to fine-tune a base model on your own data. We will explain three steps:

  1. Convert your data to a LeRobot dataset (which we use for training)
  2. Defining training configs and running training
  3. Spinning up a policy server and running inference

1. Convert your data to a LeRobot dataset

We provide a minimal example script for converting Libero data to a LeRobot dataset in examples/libero/convert_libero_data_to_lerobot.py. You can easily modify it to convert your own data! You can download the raw Libero dataset from here, and run the script with:

uv run examples/libero/convert_libero_data_to_lerobot.py --data_dir /path/to/your/libero/data

2. Defining training configs and running training

To fine-tune a base model on your own data, you need to define configs for data processing and training. We provide example configs with detailed comments for Libero below, which you can modify for your own dataset:

  • LiberoInputs and LiberoOutputs: Defines the data mapping from the Libero environment to the model and vice versa. Will be used for both, training and inference.

Online Dataset

[listener] Started on 127.0.0.1:50111
[features] Expecting frame keys: ['image', 'wrist_image', 'state', 'actions', 'timestamp', 'frame_index']
[episodes] Selected indices: [0, 1]
Generating train split: 152 examples [00:00, 8929.55 examples/s]

Script for testing data transmission

python scripts/online_stream_two_episodes.py \
  --base-repo /gscratch/socialrl/sriyash/openpi/data/task_44 \
  --online-dir /tmp/online_test_task_44 \
  --host 127.0.0.1 \
  --port 50111 \
  --delay 0.02

Script for running online finetuning

XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 python scripts/test_online_dataset_training.py \
    --config-name pi0_libero_low_mem_finetune_online_sriyash \
    --source-online-dir /gscratch/scrubbed/sriyash/openpi/data/task_44_src \
    --online-dir /tmp/openpi-online-task44 \
    --exp-name online-test \
    --num-train-steps 200 \
    --copy-delay 0.2 \
    --online-host 127.0.0.1 \
    --online-port 9999

About

No description, website, or topics provided.

Resources

Readme
License

Contributing

Contributing
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages