PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning

Angel Villar-Corrales, Sven Behnke · 2025

Read PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning by Angel Villar-Corrales, Sven Behnke (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-07600.txt.

Description

Angel Villar-Corrales and Sven Behnke present PlaySlot, a video model that learns how objects move and what hidden actions drive them, all from unlabeled clips. Their paper appeared in two thousand twenty-five. Robots need to predict how scenes unfold to plan their moves. Most predictors demand videos labeled with exact actions, which are rare.

PlaySlot splits each scene into objects and infers a latent action for each step. Those actions can come from the video itself, from a user, or from a learned policy. Conditioned on them, the model rolls out several possible futures. It beat both random-outcome and object-based rivals at prediction across settings. The inferred actions also taught robot behaviors from plain demonstration videos, with few samples. Videos and code are shared on the project page.

Catalog record

CREATORAngel Villar-Corrales, Sven Behnke
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE77 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-07600

Browse the full library · About the Tower of Babel