Hemlock
Hemlock (operating as Muscle Memory) is a data foundry for physical AI, producing force-annotated datasets that capture skilled human hand movements. Its proprietary Bracelet, an EMG wearable, records muscle activity synced with egocentric video to provide robotics developers the multimodal training data needed for dexterous manipulation tasks. The company has demonstrated its datasets in case studies with leading AI models including π0.5 and Gemini 4 Argon.
- Robotics
- Artificial Intelligence
- Hardware
- Manufacturing / Industry 4.0
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Industrial robots and physical AI systems struggle with dexterous manipulation tasks that require fine motor skills and tactile sensitivity. Standard training datasets, typically sourced from generic video or simulated environments, lack the granular contact and force information necessary for robots to learn precise, real-world manual tasks.
Solution
Muscle Memory operates as a data foundry dedicated to physical AI, producing specialized datasets that capture human hand movements with high-fidelity sensory annotations. The core of this offering is the Bracelet, a proprietary EMG wearable that records muscle activity, synced with egocentric video to provide layered contact and force data. These annotated recordings give robotics developers the rich, multimodal context needed to train systems for tasks requiring precise handling, such as assembly or surgical procedures. The company showcases its data quality through case studies with leading AI models, demonstrating real-world applicability.
Target Audience
Robotics companies, AI research labs, and manufacturers developing physical AI systems that need high-quality human demonstration data for dexterous manipulation training.
Features
- Proprietary EMG wearable (the Bracelet) that captures muscle activation signals synchronized with video
- Egocentric video recording to document the visual context and hand positioning from the worker's perspective
- Detailed annotation layer mapping contact events and force application across each recorded action
- Curated datasets tailored for specific industrial use cases in dexterous manipulation
- Demonstrated compatibility with frontier AI models including π0.5 and Gemini 4 Argon