What inspired us Robotics and world-model teams keep getting better at learning from demonstration. The missing piece we kept running into was not more models. It was usable episodes: first-person recordings of real work, cut and structured so a team can actually train on them. What problem we were trying to solve Raw hours of egocentric video are easy to collect. Training-ready episodes are harder. Teams need a clear task, a real scene, action boundaries, failure and recovery, rights and consent, and delivery scoped to their requirements. Without that, footage stays a demo reel instead of learning material. How the approach evolved We started by treating capture, labeling, and in-house pipelines as one problem with three entry points. Some teams need us to film the task. Some bring existing recordings for structure and review. Some collect in-house and want help scoping episode structure and delivery. The headline stayed the same: egocentric video in, training-ready episodes out. The work got more concrete around tasks, cuts, and what "done" means for an episode. If you are building robots, world models, or research datasets from human demonstration, tell us what your team needs to learn. Build with Wake: https://wake.inc/contact/

WAKE

Training episodes from egocentric video

Training episodes from egocentric video