Encord releases EMG-1 dataset for robot hand tracking

Encord said in a press release that it released EMG-1, an open dataset pairing forearm muscle signals with video recorded from a wearer's perspective. Developed with Wearable Devices, the dataset contains more than 300 hours of synchronized recordings from real world tasks, along with optical motion capture data.

The wristbands used to collect the data record electrical activity linked to finger movements. This can provide hand pose labels when objects or other obstacles hide a person's hands from cameras, while also capturing signals related to force and contact.

In Encord's tests, combining muscle signals with video reduced fingertip tracking error by 25% compared with video alone. It also lowered error by 35% during brief gaps in visual tracking. The release accompanies the launch of Encord Data Labs, a research division focused on data for robotics and physical AI.