Fast-WAM: Do World Action Models Need Test-time Future Imagination?
Original research
MODEL PROFILE · World-action research
Fast-WAM
Uses video-model representations to predict actions without generating full video at inference.
Research organisationTsinghua / Galaxea AI
Research approachRepresentation-only
Published year2026
BackboneWan 2.2 (5B)
Overview
Fast-WAM explores a representation-only route to faster action inference. The WAM survey describes its use of a pretrained Wan video backbone while skipping full video generation during robot action prediction.
Source record
- Fast-WAM: Do World Action Models Need Test-time Future Imagination?
- NVIDIA WAM technical synthesis (Moritz Reuss, 15 Jun 2026)
Core research identity and broad method were source-checked as of 2026-11-03. Independent benchmarking, commercial availability, weights and licensing have not been assessed.