Search papers, labs, and topics across Lattice.
This paper introduces Spatial Memory Intelligence (SMI), a novel framework that leverages an understanding model to manage long-term spatial context in world models for long-video generation. SMI addresses the challenge of increasingly complex and lengthy memory sequences by employing four coordinated operations: spatial clustering, within-cluster sparsification, action-aware retrieval, and reliability-aware filtering. Experimental results show SMI significantly improves memory sparsity, generation stability, and spatial consistency across various benchmarks and world-model architectures.
An understanding-driven approach unlocks systematic spatial-memory management for world models, dramatically improving long-video generation quality.
Long-video generation and world models have shown strong potential for interactive entertainment and embodied simulation by predicting future observations conditioned on user actions and historical memory. However, as memory sequences grow longer and their structures become increasingly complex, managing long-range spatial context becomes increasingly challenging, calling for a more intelligent and systematic memory-management strategy. Building on the advancing spatial reasoning capabilities of multimodal large language models (MLLMs) and the broader vision of unified models, we propose Spatial Memory Intelligence (SMI), the first framework to systematically employ an understanding model for spatial-memory management in long-video world models. SMI introduces four coordinated atomic operations: spatial clustering, within-cluster sparsification, action-aware retrieval, and reliability-aware filtering. Extensive experiments across multiple baselines, benchmarks, and world-model backbones demonstrate the effectiveness and generalizability of SMI, achieving comprehensive improvements in memory sparsity, generation stability, and spatial consistency.
