XSquare Robot unveils WALL-SS: long-horizon world model with 0.93 sim-real correlation
chris_j_paxton · x · 2026-08-28
XSquare Robot introduces WALL-SS, a next-scale autoregressive world model for action-controllable, long-horizon robotic simulation. It targets three core challenges: making predicted futures genuinely follow action commands, enabling stable rollouts up to 60 seconds, and aligning virtual predictions with real-world robot execution.
The model generates future video from coarse structure to fine physical detail while aligning robot actions across time, visual scales, and camera views, so generated motion follows prescribed controls more faithfully and stays coherent over extended rollouts.
In closed-loop tests across six manipulation tasks and 600 matched rollouts, simulated and real-world policy success rates agreed with MAE 0.062 and a Pearson correlation of 0.93. Project page and paper are public.
Related event: XSquare Robot Unveils WALL-SS World Model for Long-Horizon Robot Simulation(2 posts)→
More from Embodied
- VR Pro Controllers vs. Default: World Space SLAM vs. IR Tracking — Darpinian · 2026-08-28
- Vision Pro controller tracking: Pro controllers do world-space SLAM directly, key for teleop — Darpinian · 2026-08-28
- AME-2 paper accepted: Agile legged locomotion via attention maps — ChongZzZhang · 2026-08-28
- Philips gets $34 million U.S. grant to develop AI robots for stroke treatment — boppinmule · 2026-08-28
- Founder declines $65M offer to protect Indigenous language AI — danielleboyerr · 2026-08-28
- RoboCurve launches as a neutral third party for real-world robotics evals — ycombinator · 2026-08-28