# IMPULSE: Push First, Then Predict in V-JEPA 2 Feature Space

DevFeed: [IMPULSE: Push First, Then Predict in V-JEPA 2 Feature Space](<https://devfeed.tech/articles/impulse-push-first-then-predict-in-v-jepa-2-feature-space-93219.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.23968>)

Author: Tingan Jin

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Feature Engineering](<https://devfeed.tech/topics/feature-engineering.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [bayesian](<https://devfeed.tech/tags/bayesian.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [error](<https://devfeed.tech/tags/error.md>), [memory](<https://devfeed.tech/tags/memory.md>), [motion](<https://devfeed.tech/tags/motion.md>), [physics](<https://devfeed.tech/tags/physics.md>), [random](<https://devfeed.tech/tags/random.md>), [video](<https://devfeed.tech/tags/video.md>), [world-models](<https://devfeed.tech/tags/world-models.md>)

## AI overview

IMPULSE improves V-JEPA 2 feature forecasts by letting a predictor use experience from three calibration pushes. A Bayesian contact memory condenses the pushes into eight interpretable numbers; predicted object pose carries this information into a frozen-feature predictor, and a random-Fourier pose head helps the predictor use pose. In a held-out confirmation on 512 fresh objects, memory reduced 1-second forecast error by 11.6-12.8% versus the same predictor without experience, across all eight training seeds, with a 95% interval excluding zero. Gains also appeared on 128 objects with out-of-range physics. The Fourier head produced roughly 2.8-3.2 times the gain of perfect memory over plain pose tokens; raw push measurements through the same pathway gave similar gains, so Bayesian summarization was not necessary. Improvement was absent at contact onset and grew afterward. These findings are reported in the supplied abstract.

## Source excerpt

arXiv:2610.23968v1 Announce Type: cross Abstract: Latent world models such as V-JEPA 2 forecast in the feature space of a video encoder, but one look at an object does not reveal how hard it is to rotate, how much it bounces or how much it slides. People find out by trying: a few pokes reveal how an object responds. IMPULSE, named after the impulse response it infers, gives a predictor in frozen V-JEPA 2 feature space the same ability. A Bayesian contact memory summarizes three short calibration pushes, a predicted object pose carries it into the predictor, and a lightweight random-Fourier pose head lets the predictor use the pose. On 512 fresh objects, in a pre-specified held-out confirmation with models sealed before the data existed, the memory lowers the 1-second feature-forecast error by 11.6-12.8% relative to the same predictor without experience, for all 8 training seeds, with a 95% seed-and-object interval excluding zero. The gain also holds for 128 objects with out-of-range physics (11.7-14.0%), and the full extension beats a plain predictor by 15.5-16.6%. An information-ladder argument, exact for ideal predictors, shows that physics reaches the features only through motion, so we measure how much the pathway lets pose help. On fresh objects, the Fourier head roughly triples the gain of a perfect memory over plain pose tokens (2.8-3.2x). Raw push measurements fed through the same pathway give a similar gain, so the Bayesian summary is not required; it condenses the trials into eight interpretable numbers. The gain is absent at contact onset and grows afterwards.