INTACT · July 30, 2026
INTACTPaper availableCode and weights
Default checkpoint · INTACT · Single-task E1
Participating organizationsRoboParty
Participating organizationsTsinghua UniversityParticipating organizationsZhejiang UniversityVersion scope and roles for these associations await verification.
INTACT is an end-to-end JEPA world model that learns from reward-free offline trajectories labeled with actions. It encodes current and goal RGB frames and uses their latent difference to generate action chunks directly. Evaluation covers four simulated visual control tasks.
Paper available
Author order: Original order
Publisher not provided · Documentation
Publication date unknown · Accessed: 2026-11-21
Publisher not provided · Source
Publication date unknown · Accessed: 2026-11-21
Publisher not provided · Code repository
Publication date unknown · Accessed: 2026-11-21
Publisher not provided · Code repository
Publication date unknown · Accessed: 2026-11-21