Hebbian Robotics Launches HFlow Open-Source SDK for Robotics Data Pipelines

Introducing HFlow SDK

Hebbian Robotics has released HFlow, an open-source Software Development Kit (SDK) aimed at streamlining multimodal data pipelines in robotics and physical AI. The SDK is designed to make data tooling and practices, typically found in large robotics organizations, available to teams of all sizes.

Addressing Data Bottlenecks

The development of HFlow stems from the observation that data processing is a significant bottleneck in robotics. Robotics data corpuses often combine various data types such as video, state information, actions, timestamps, and metadata from multiple recording systems. Challenges include quality control issues like frozen cameras, out-of-sync streams, missing data topics, and duplicate recordings. As data grows, fragmented scripts complicate auditing, reproduction, and understanding of processing steps.

HFlow's Core Functionality

HFlow provides built-in checks and allows users to integrate custom transformations, checks, labels, and enrichments. The SDK manages the orchestration, storage, versioning, and curation of these processing steps. It records the provenance of each processed episode, visualizes pipelines as graphs, and stores metadata and quality evidence in a queryable catalog. This enables tracing of output production, monitoring of stages, and investigation of data without loading raw recordings.

Data Format and Lifecycle

HFlow utilizes MCAP as its v1 input and output boundary, chosen for its efficiency in storing and serving synchronized video, state, action, and other time-series streams. Data from various collection systems, including human-worn cameras, teleoperated robots, and autonomous policies, can feed into the pipeline once represented as a supported MCAP episode. The data lifecycle within HFlow progresses through collection, ingestion, curation, and delivery stages.

Current Status and Community Engagement

HFlow is currently in a pre-v1 state, with its core lifecycle functionality operational end-to-end. It is available for local testing. Hebbian Robotics encourages community contributions, aiming to foster an open-source environment for robotics development that does not require robot hardware to participate.