Releases·Tool Release
Hebbian Robotics (YC S26) launches HFlow SDK for robotics data pipelines
Hebbian Robotics, founded by Brandon and Kingston (YC S26), launched HFlow on Launch HN. The open-source SDK turns multimodal robot and human-operator recordings into standardized, quality-checked episodes and queryable dataset manifests. It targets teams whose robotics data pipelines begin as ad-hoc scripts.
56 engagement1 sourceMon, Aug 31, 2026, 03:02 PM
AI Overview
- HFlow processes synchronized video, joint states, actions, timestamps, and metadata from a single recording together
- Output is standardized, quality-checked episodes plus queryable dataset manifests
- Source code is on GitHub (Hebbian-Robotics/hflow); a YouTube demo shows the tool in action
- Founders frame the problem as pipelines that start as scripts: one transcodes video, etc.
Entities
Related
Developments·Benchmark
HFlow evaluation finds open-weight VLMs closely match Gemini 2.5 Flash on egocentric data
Hebbian Robotics used its open-source HFlow tool to rerun Build AI's Egocentric-10K evaluation across recent open-weight VLMs, keeping the same prompts and dataset and varying only the model. Agreement with the Gemini 2.5 Flash baseline reached 91% for several open models, suggesting small open-weight VLMs are viable for egocentric video processing tasks like hand visibility and active manipulation detection.
Hebbian Robotics used its open-source HFlow tool to rerun Build AI's Egocentric-10K evaluation across recent open-weight VLMs, keeping the same prompts and dataset and varying only the model. Agreement with the Gemini 2.5 Flash baseline reached 91% for several open models, suggesting small open-weight VLMs are viable for egocentric video processing tasks like hand visibility and active manipulation detection.
22 engagement1 sourceredditSun, Aug 30, 2026, 07:40 PM
Releases·Tool Release
Grabette: open system for recording robot manipulation data
Grabette is an open system designed to record robot manipulation data, addressing the shortage of large, diverse real-world datasets for robot learning. It converts raw recordings into LeRobot datasets, aiming to reduce the cost and complexity of data collection.
Grabette is an open system designed to record robot manipulation data, addressing the shortage of large, diverse real-world datasets for robot learning. It converts raw recordings into LeRobot datasets, aiming to reduce the cost and complexity of data collection.
0 engagement1 sourcerssTue, Jul 21, 2026, 12:00 AM
Products·launched
Local web app for curating SO-101 datasets for VLA/ACT training
Dataset Assembly Studio is a local web app that helps robotics researchers curate and combine SO-101 datasets from different versions, framerates, and camera setups for training models like Lingbot-va and Dreamzero. It addresses the problem of messy, repetitive datasets with generic task descriptions by providing tools to filter, balance, and assemble them into usable training sets. LLMs are used to generate or refine task descriptions and possibly to summarize or deduplicate data.
Dataset Assembly Studio is a local web app that helps robotics researchers curate and combine SO-101 datasets from different versions, framerates, and camera setups for training models like Lingbot-va and Dreamzero. It addresses the problem of messy, repetitive datasets with generic task descriptions by providing tools to filter, balance, and assemble them into usable training sets. LLMs are used to generate or refine task descriptions and possibly to summarize or deduplicate data.
3 engagement1 sourceredditThu, Aug 13, 2026, 07:28 AM
Releases·Tool Releasefeatured
Figure.AI releases Index, a 16M-video robot dataset with paid data collection
Figure.AI released Index, a robot dataset containing 16 million videos, described as the largest and most diverse of its kind. The company is now paying people to record their daily tasks to expand the dataset, which is intended to improve robot learning.
Figure.AI released Index, a robot dataset containing 16 million videos, described as the largest and most diverse of its kind. The company is now paying people to record their daily tasks to expand the dataset, which is intended to improve robot learning.
410 engagement1 sourceredditTue, Aug 25, 2026, 07:20 PM
Developments·Paper
Pegasus framework bridges embodiment gap by translating human videos into robot training data
Pegasus is a low-resource framework that converts human demonstration videos into robot-learnable data using a graph-based intermediate representation. It extracts a Task Graph from human videos, then transforms it through Affordance and Constraint mapping to generate physically grounded robot experiences. This addresses the key bottleneck of data scarcity for embodied AI without requiring expensive robot-specific data collection.
Pegasus is a low-resource framework that converts human demonstration videos into robot-learnable data using a graph-based intermediate representation. It extracts a Task Graph from human videos, then transforms it through Affordance and Constraint mapping to generate physically grounded robot experiences. This addresses the key bottleneck of data scarcity for embodied AI without requiring expensive robot-specific data collection.
0 engagement1 sourcearxivWed, Jul 29, 2026, 01:35 PM