AI TrackerFrontier desk · AEST
Back to news
normalaitech

T2-z: Test-Time Optimization of Environment Latents for Robotic World Models

arXiv cs.AI · 16 Oct 2026, 03:59 AEST

T2-Z introduces a method for optimizing environment latents in robotic world models to improve predictions. The approach allows robots to adapt to unseen environmental configurations during test time, significantly enhancing performance metrics.

Key facts

Read aloud

T2-Z introduces a method for optimizing environment latents in robotic world models to improve predictions. The approach allows robots to adapt to unseen environmental configurations during test time, significantly enhancing performance metrics. Predicting how the world responds to an action requires more than what can be seen: robots must account for environmental factors that are neither directly observable nor explicitly specified. We introduce T2-Z for test-time optimization of environment latents, which augments a robotic world model with a learnable environment latent to condition world predictions. At test time, the world model remains frozen and a newly initialized environment latent is optimized from one or more test-time interactions, adapting the model to previously unseen environmental configurations. Across seven simulated settings and four real-world robotic tasks, T2-Z reduces object prediction error metric from fifteen point two zero to seven point zero zero and raises action success from thirty three point seven three percent to seventy two point nine six percent. Ablations further show that T2-Z can efficiently infer environment latents from as little as a single test-time interaction trajectory.

Sources in this cluster