Generalization beyond the Epona backbone

Establish whether SIMVTA generalizes to world action model architectures other than the Epona backbone.

Background

The experiments validate SIMVTA using the Epona world action model backbone. Because the framework jointly updates trajectory-planning and video-prediction components through shared representations, its effectiveness may depend on architectural properties specific to Epona.

The authors explicitly leave validation on alternative world action model architectures unresolved, making cross-architecture generalization an open direction.

References

Generalization to alternative WAM architectures is left for future work.

— SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action Models  (2610.13781 - Jing et al., 9 Oct 2026) in Appendix G, “Limitations,” subsection “Architecture and Training Cost”