Temporal Forcing: 4D Representation Alignment for Vision-Language-Action Models
2026-08-31 • Robotics
Robotics
AI summaryⓘ
The authors improve how robots understand and act in changing 3D environments by teaching their model to remember what happened before, not just what is happening now. They add a new way for the model to keep track of time and align this information with advanced 4D geometric data that shows how scenes evolve. This helps the robot better handle tasks that take longer and involve looking very similar through time. Their method, called Temporal Forcing, performs better on standard tests and difficult physical tasks than older models.
vision-language-action3D scene geometrytemporal information4D representationlatent representationsfoundation modellong-horizon manipulationobservation aliasingdynamic environmentsrobot manipulation
Authors
Xingyu Ding, Yuzhong Zhao, Chunhai Zhao, Yinghuan Shi, Chaoyang Zhao, Yifan Zhang
Abstract
Recent vision-language-action (VLA) methods improve manipulation performance by aligning their representations with 3D scene geometry. However, these methods often struggle with long-horizon manipulation and observation aliasing between visually similar states due to a lack of temporal information: the 3D scene geometry captures only the current state, rather than how it has evolved over time. To resolve this, we present Temporal Forcing, a 4D representation alignment method for VLA models. Specifically, we first introduce a history pathway that enables a vanilla VLA model to summarize observation history into temporally aware latent representations. Then, the latent representations are aligned with the geometric features extracted by a pretrained 4D foundation model, which captures the evolving 3D world through temporally consistent geometric representations, enabling a deeper understanding of dynamic environments. Temporal Forcing reaches 98.8% on LIBERO, outperforming its base model by 2.2 points. On a physical hidden-placement task, it raises full-task success from 20.0% to 43.3%. Code will be publicly available.