Researchers introduce KinoGen, a framework for synthesizing customizable humanoid loco-manipulation references. Instead of relying solely on unconstrained video generation, KinoGen conditions a pretrained video generator on calibrated renders, providing kinematic grounding.

According to the paper's abstract on arXiv, this grounding is what lets the framework produce references that are customizable. The method is aimed at humanoid loco-manipulation—tasks that combine locomotion and manipulation—where reference videos are often scarce or hard to capture.

Because the source is a single abstract, details about evaluation, baselines, and limitations are not available in this material.