# Kavana

> Kavana captures hand pose, force and intent with Mudra™ neuromuscular (surface EMG) wristbands. It is the human layer for physical-AI data collection and for training models of dexterous robotic hands. Kavana is the physical AI unit of Wearable Devices Ltd. (Nasdaq: WLDS).

Key facts:
- Core idea: biopotentials reveal what cameras cannot. Vision-only hand capture fails on occlusion, reflections and glare, motion blur and hands leaving the field of view, and it cannot see force at all. Wrist EMG, IMU and PPG fill those gaps.
- Kavana fuses its EMG wristband with any egocentric camera array, with no gloves, no line-of-sight dependency and no extra rigs. In Kavana's tests, adding EMG to an egocentric camera halved hand pose error, from 9.0° to 4.2° mean absolute error against motion capture.
- EMG correlates with grip and contact force in Newtons, so trained models can estimate force from the wrist.
- Track record: 13 patents (7 granted, 6 pending) in surface EMG and neural input, more than 12 years of surface EMG R&D, and 4 generations of consumer hardware.
- Hardware: Mudra Ultimate is the research-grade capture band for physical-AI companies collecting data at scale (8 EMG channels, 500–4000 Hz, Wi-Fi streaming, multi-sensor sync, 43 g). Mudra Pro is the light everyday band for research labs and academic groups (3 EMG channels, up to 6400 Hz, 38 g). Both have 24-bit EMG, a 6-DoF IMU, PPG, USB-C and BLE, and optional on-device SD recording.
- Software: Python and C++ SDKs for Windows, macOS and Linux, fully agentic, with Lab Streaming Layer (LSL) support. Open source repository: Indriya on GitHub.
- The Kavana lab records people doing real manipulation tasks. It combines a 12-camera OptiTrack motion capture volume, Mudra EMG, egocentric RGB or RGB-D video, force sensors and tactile gloves, all on one LSL clock. The output is multimodal datasets for joint-angle estimation, computer-vision uplift and applied-force estimation.
- Working with Kavana is a three-step data program: Scope (one call to define the task, the modalities and the target metric), Pilot (a small synchronized recording you train on) and Scale (more subjects, sites and session hours if the pilot moves the metric).

## Platform

- [Home: hardware, specs and data programs](https://www.kavana.tech/): Why EMG beats vision-only capture, the camera and EMG fusion results, Mudra Pro and Mudra Ultimate specs, and how to start a data program
- [Our lab](https://www.kavana.tech/our-lab): The capture rig, synchronization layer and multimodal datasets for dexterous manipulation
- [Start a data program](https://www.kavana.tech/#contact): Contact form for a scoping call, a pilot or scaling an existing program. Replies within two working days

## Research

- [Blog](https://www.kavana.tech/blog): Research results, lab methods and product notes on capturing hand pose, force and intent with Mudra neuromuscular wristbands
- [Research sources](https://www.kavana.tech/tactile-gap-sources): 184 cited papers, including 68 vision-language-action models that add touch, force or other sensing to vision and 116 papers estimating hand pose from surface EMG
- [Indriya repository](https://github.com/kavana-tech/Indriya): Open source code and SDK resources for Mudra wristbands

## Company

- [About us](https://www.kavana.tech/about): The interdisciplinary team and mission behind Kavana
- [Press kit](https://www.kavana.tech/press-kit): Logos, product images, lab images, videos and company background
- [Investor relations](https://www.wearabledevices.co.il/): Wearable Devices Ltd. (Nasdaq: WLDS), Kavana's parent company

## Optional

- [Privacy policy](https://www.kavana.tech/privacy-policy)
