ActionSplice edits actions in video world models without replaying evaluations
TexasAMUniversity · hf · 2026-09-14
Texas A&M University released ActionSplice on Hugging Face, an in-flight action editing method for interactive world models. It introduces counterfactual state transport to splice revised actions into chunk-autoregressive video world models without replaying completed evaluations, improving both fidelity and speed.
More from Research
- PufferLib 5.0 Hits 60M Steps/Sec on a Single RTX 5090: How 5 Degrees of Parallelism Work — jsuarez · 2026-09-14
- PufferLib 5.0: Five Coolest RL Results, from Multitask Drone Policies to Selfplay Robocode Champions — jsuarez · 2026-09-14
- Ocean adds ~60 free open-source RL environments with browser-runnable baselines — jsuarez · 2026-09-14
- PufferLib Ocean now offers ~60 RL environments with browser-runnable baselines — jsuarez · 2026-09-14
- PufferLib 5.0 is ~5x faster than 4.0, peak throughput over 60M SPS — jsuarez · 2026-09-14
- PufferLib team showcases 5 new results with playable agents online — jsuarez · 2026-09-14