neuracore 1.6.3

Neuracore Client Library

Neuracore Python Client

Neuracore is a powerful robotics and machine learning client library for seamless robot data collection, model deployment, and interaction.

Features

• Easy robot initialization and connection
• Streaming data logging
• Model endpoint management
• Local and remote model support
• Flexible dataset creation
• Open source training infrastructure with Hydra configuration

Installation

pip install neuracore

For training and ML development:

pip install neuracore[ml]

Quick Start

Ensure you have an account at neuracore.app

Authentication

import neuracore as nc

# This will save your API key locally
nc.login()

Robot Connection

# Connect to a robot
nc.connect_robot(
    robot_name="MyRobot", 
    urdf_path="/path/to/robot.urdf"
)

You can also upload MuJoCo MJCF rather than URDF. For that, ensure you install extra dependencies: pip install neuracore[mjcf].

nc.connect_robot(
    robot_name="MyRobot", 
    mjcf_path="/path/to/robot.xml"
)

Data Logging

# Log joint positions
nc.log_joint_positions({
    'joint1': 0.5, 
    'joint2': -0.3
})

# Log RGB camera image
nc.log_rgb("top_camera", image_array)

Open Source Training

Neuracore includes a powerful open-source training infrastructure built with Hydra for configuration management. Train your own robot learning algorithms locally rather than using our cloud training service.

Training Setup

The training system is located in neuracore/ml/ and includes:

neuracore/
  ml/
    train.py              # Main training script
    config/               # Hydra configuration files
      config.yaml         # Main configuration
      algorithm/          # Algorithm-specific configs
        diffusion_policy.yaml
        act.yaml
        simple_vla.yaml
        ...
      training/           # Training-specific configs
      dataset/            # Dataset-specific configs
    algorithms/           # Built-in algorithms
    datasets/             # Dataset implementations
    trainers/             # Distributed training utilities
    utils/                # Training utilities

Quick Training Examples

Local Development Training

# Train with built-in Diffusion Policy
python -m neuracore.ml.train algorithm=diffusion_policy dataset_name="my_dataset"

# Train with ACT algorithm and custom parameters
python -m neuracore.ml.train algorithm=act algorithm.lr=5e-4 algorithm.hidden_dim=1024 dataset_name="my_dataset"

# Auto-tune batch size
pythonv neuracore.ml.train algorithm=diffusion_policy batch_size=auto dataset_name="my_dataset"

# Hyperparameter sweeps
python -m neuracore.ml.train --multirun algorithm=cnnmlp algorithm.lr=1e-4,5e-4,1e-3 algorithm.hidden_dim=256,512,1024 dataset_name="my_dataset"

Configuration Management

Neuracore uses Hydra for flexible configuration management:

# config/config.yaml
defaults:
  - algorithm: diffusion_policy
  - training: default
  - dataset: default

# Core parameters
epochs: 100
batch_size: "auto"
seed: 42

# Data types
input_data_types:
  - "joint_positions"
  - "rgb_image"
output_data_types:
  - "joint_target_positions"

Custom Algorithm Development

Create your own algorithms by inheriting from NeuracoreModel:

from neuracore.ml import NeuracoreModel
from neuracore.core.nc_types import DataType

class MyCustomAlgorithm(NeuracoreModel):
    def __init__(self, model_init_description, **kwargs):
        super().__init__(model_init_description)
        # Your initialization here
        
    def forward(self, batch):
        # Your inference logic
        pass
        
    def training_step(self, batch):
        # Your training logic
        pass
        
    @staticmethod
    def get_supported_input_data_types():
        return [DataType.JOINT_POSITIONS, DataType.RGB_IMAGE]
        
    @staticmethod
    def get_supported_output_data_types():
        return [DataType.JOINT_TARGET_POSITIONS]

Features

• Distributed Training: Multi-GPU support with PyTorch DDP
• Automatic Batch Size Tuning: Find optimal batch sizes automatically
• Memory Monitoring: Prevent OOM errors with built-in memory monitoring
• TensorBoard Integration: Comprehensive logging and visualization
• Checkpoint Management: Automatic saving and resuming
• Cloud Integration: Seamless integration with Neuracore SaaS platform
• Flexible Data Types: Support for multi-modal robot data (images, joint states, language)

Algorithm Validation

Validate your algorithms before training:

neuracore-validate /path/to/your/algorithm

Documentation

• Examples
• Creating Custom Algorithms
• Limitations

Development

To set up for development:

git clone https://github.com/neuraco/neuracore
cd neuracore
pip install -e .[dev,ml]

Environment Variables

A few environment variables effect how this library operates, they are generally prefixed with NEURACORE_ and are case insensitive

Variable	Function	Valid Values	Default Value
NEURACORE_REMOTE_RECORDING_TRIGGER_ENABLED	Allows you to disable other machines starting a recording when logging. this does not affect the live data	true/false	true
NEURACORE_LIVE_DATA_ENABLED	Allows you to disable the streaming of data for live visualizations from this node. This does not affect recording.	true/false	true
NEURACORE_API_URL	The base url used to contact the neuracore platform.	A url e.g. https://api.neuracore.app/api	https://api.neuracore.app/api

Testing

export NEURACORE_API_URL=http://localhost:8000/api
pytest tests/

Contributing

Contributions are welcome!