The dataset comprises nearly 20,000 paired samples spanning thousands of motions. Using multimodal contrastive learning, the team showed that non-invasive EEG embeddings can align with text, video and motion representations to decode high-level semantics. EMMM couples an EEG encoder with a motion decoder, and experiments indicate it generates coherent, realistic motion sequences from non-invasive brain activity. To the researchers’ knowledge, this is the first work to generate diverse full-body human motions from non-invasive brain signals, opening a direction toward generative and open-vocabulary motor BCIs.
EEG2MOTION: Full-Body Motion From Brain Signals
Summary
Researchers debut EEG2MOTION, billed as the first EEG-motion-text dataset for human motion synthesis, with nearly 20,000 paired samples across thousands of motions, plus a generative framework (EMMM) that couples an EEG encoder with a motion decoder to synthesize coherent full-body motions from non-invasive brain activity. Multimodal contrastive learning aligns non-invasive EEG embeddings with text, video and motion representations to decode high-level semantics. The team says it is the first work to generate diverse whole-body motions from non-invasive brain signals.
Why it matters
Open-vocabulary full-body motion synthesis is a frontier beyond single-command BCI control; EEG2MOTION supplies the first dataset and generative baseline for it, so the preprint matters even before peer review.
Compiled by BCIwiki from public sources
Sources · 1
arxiv.org 2026-08-14