RLWRLD Lands 5 CoRL 2026 Papers; Human-Video Pretraining Lifts VLA Success to 80.3%

chris_j_paxton · x · 2026-09-12

Robotics startup RLWRLD announced 5 papers accepted to CoRL 2026, headlined by HuRo:

HuRo builds a "robotization" pipeline converting heterogeneous human videos into robot-aligned observations and action trajectories, yielding 630K episodes and 142M frames for VLA pretraining
Scaling this pretraining lifts overall task completion from 51.5% to 80.3% across four real-world manipulation tasks, and OOD completion under spatial/visual shifts from 34.9% to 72.2%
Ablations show visual robotization improves OOD robustness, and end-to-end pretraining with retargeted actions beats visual-only transfer; code and data are released
Other accepted papers cover folding tactile/torque feedback into VLA action streams via cross-modal self-attention, and learning invertible latent decompositions of collision-free configuration space for path planning

More from Embodied