Yu Liu

Robot world models can now achieve 6.25% interaction defect rates, a 7-fold reduction, by training on counterfactual futures and a novel embodied video reward.

Predicting task-relevant latent futures, not raw pixels, unlocks efficient and effective long-horizon VLA control, outperforming reactive policies by over 11 percentage points.

A single forward pass through an atomistic foundation model can predict material stability with 30x less computation than a full relaxation, while providing calibrated probabilities and a clear rule for when to "think slowly."

Non-centrosymmetric Ce$_4$Ge$_7$ hosts an intrinsic Karplus-Luttinger anomalous Hall effect alongside a pressure-tuned quantum critical point at 10.2 GPa, exposing a rare intersection of strong correlations and topological transport.

Localized $4f$ honeycomb ferromagnetism directly couples to non-trivial Weyl topology in Ce$_2$Zn$_6$Ge$_3$, establishing a rare platform with accessible Weyl nodes positioned near the Fermi energy.

Long-horizon agent RL frequently stalls because fixed rewards cannot track shifting bottlenecks—co-evolving evaluation rubrics with targeted exploration skills directly breaks the repetitive failure loops of current policies.

Deep-space edge AI no longer requires heavy radiation shielding: this bare-die antiferromagnetic processor executes direct analog neural classification at 0.2 fJ per operation while surviving 55 T magnetic fields and 1.5 Mrad radiation doses.

MachEmbodied-U0, a unified embodied foundation model connecting understanding and generation experts through a Mixture-of-Transformers architecture, is presented, a unified embodied foundation model connecting understanding and generation experts through a Mixture-of-Transformers architecture that combines competitive downstream control performance with transferable task-grounding and visual-dynamic capabilities across simulation and the real world.

PersonaPath, a benchmark that pairs 2,000 fine-grained learner personas with a hierarchical knowledge graph of 347 textbooks, 1,751 units, and 4,092 concepts across 77 subjects, is introduced and results show that even the strongest LLM reaches only a 29.5% final pass rate in Basic Education.