Mengzhan Liufu | Searcharxiv

arXiv2026

Out of Sight, Out of Mind? Evaluating State Evolution in Video World Models

Evolutions in the world, such as water pouring or ice melting, happen regardless of being observed. Video world models generate "worlds" via 2D frame observations. Can these generated "worlds" evolve regardless of observation? To probe this question, we design a benchmark to evaluate whether video world models can decouple state evolution from observation. Our benchmark, STEVO-Bench, applies observation control to evolving processes via instructions of occluder insertion, turning off the light, or specifying camera "lookaway" trajectories. By evaluating video models with and without camera control for a diverse set of naturally-occurring evolutions, we expose their limitations in decoupling state evolution from observation. STEVO-Bench proposes an evaluation protocol to automatically detect and disentangle failure modes of video world models across key aspects of natural state evolution. Analysis of STEVO-Bench results provide new insight into potential data and architecture bias of present-day video world models. Project website: https://glab-caltech.github.io/STEVOBench/. Blog: https://ziqi-ma.github.io/blog/2026/outofsight/

cs.CV↗

arXiv2020

Intelligent Online Food Delivery System: A Dynamic Model to Generate Delivery Strategy and Tip Advice

Due to the rapid development of online food ordering platforms and rocketing growth of demand, the market is about to saturate soon, and the future trend is to seek efficient utilization of resources. Specifically speaking, food company must have a reliable algorithm to help them produce efficient delivery strategies; individual customers need planning for their decision making in this field. For example, when customers add tip to their order with the sake of controlling or reducing latency. However, few customers know how much tip is enough to reach their desired latency. Therefore, in our paper, we establish a dynamic model to generate delivery strategy for companies and tip advice for customers. We believe that the system we design is more efficient than the currently primitive system. We simulate the delivery process and generate delivery strategies using genetic annealing because it can approach a near optimal solution. High-quality delivery queue ensures that those orders can be delivered within an acceptable amount of time. Next, we construct regressions to find out relationships between multiple factors and latency and then generate the advisory amount of tip. Finally, we plug in those values and desired waiting time, getting the advisory tip price. Multiple indexes suggest that our regression results are accurate and reliable.

stat.AP↗