Search papers, labs, and topics across Lattice.
8
0
12
0
This work proposes DIDO, which distills the converged dynamics of a multi-step video model into a single denoising step and achieves an average success rate of 99.0\% on LIBERO, 76.6\% on LIBERO-Plus, and 92.0\% on RoboTwin, while also demonstrating effective transfer to long-horizon and generalization tasks in real-world robotic manipulation.
Achieving state-of-the-art pose-free novel view synthesis by reusing a single 3D scene across multiple complementary roles transforms how we approach 3D reconstruction and rendering.
Fine-grained metrics reveal that robots can recover from failures more effectively than previously thought, reshaping our understanding of their capabilities.
OASE enables agents to adapt effectively in constantly evolving environments by selecting skill revisions based on historical opponent strategies, rather than outdated references.
Orca's unified world latent space enables superior performance in diverse tasks, outperforming specialized models with a single framework.
MLLMs struggle to convert visual evidence into context-specific actions, with performance plummeting by over 44% in certain scenarios.
RotMoLE's rotational gating unlocks more representational power from low-rank MoE architectures, even when expert diversity is limited.
Grounding RAG in historical texts demands more than semantic similarity; ChunQiuTR reveals the critical importance of temporal consistency, where even plausible evidence can be rendered invalid by subtle time-key mismatches.
