Axonnic
Axonnic provides an EMG sleeve that captures muscle activation from workers' forearms, enabling manufacturers to automate dexterous, force-sensitive tasks that traditional vision systems cannot handle. The system records expert demonstrations—combining EMG, video, and hand pose—to train robots that then operate on their own sensors without the sleeve. The sleeve tracks individual finger movements with a 4.58° average joint-angle error and is currently deployed in factory pilots for force-sensitive sewing.
- Manufacturing / Industry 4.0
- Robotics
- Industrial Automation
- Artificial Intelligence
- Hardware
- Construction Technology
Funding
Founders
Founder details are not available yet.
Product
Problem
Manufacturers struggle to automate dexterous, force-sensitive tasks like sewing, delicate assembly, and tool use because traditional vision systems only capture hand position, not the physical effort and muscle activation required to perform the work. This limits automation to rigid, predictable tasks and leaves skilled manual labor dependent on human workers.
Solution
Axonnic provides an EMG sleeve that records muscle activation from workers' forearms, capturing physical information that cameras miss. The system combines EMG, video, and hand tracking to build rich datasets of how experts actually perform tasks, then uses that data to train robots. At deployment, the robot runs on its own sensors and never wears the sleeve—the EMG guides training, then drops away. Each deployment produces physical-interaction data that improves the next automation project, creating a compounding dataset advantage.
Target Audience
Primary customers are manufacturers automating dexterous tasks in sewing, delicate assembly, machine tending, and construction, as well as robotics and AI labs seeking generalized datasets of human physical interaction.
Features
- EMG sleeve measures muscle activation and inferred effort, capturing the preparation before contact that video alone cannot reveal
- Hand remains completely unobstructed during recording, unlike glove-based systems
- Combines EMG, video, and hand pose into aligned datasets for robotics and AI labs
- Robot requires no EMG at inference time, running on its own sensors after training
- Validated in factories and construction sites, including delicate, force-sensitive sewing and work by electricians, plumbers, and demolition crews
- Each deployment generates proprietary physical-interaction data that improves subsequent training
- 64-channel sleeve tracks individual finger movements with 4.58° average joint-angle error and 50 predictions per second
- Model outperforms Meta's on their own benchmark, achieving lower error for all 20 held-out people