Poly-Autoregressive Modeling for Interacting Entities

Neerja Thakkar, Jathushan Rajasegaran, Tara Sadjadpour, Shiry Ginosar, Jitendra Malik

Keywords: autoregressive prediction; multi-agent interaction

TL;DR: We present a simple poly-autoregressive modeling framework that predicts an agent's future behavior by considering the effects that other interacting agents and entities have on them.

Abstract:

We present a simple framework that predicts an agent's future behavior by considering the effects that other interacting agents and entities have on them. We propose to model behavior as a sequence of tokens, each representing the state of an agent at a specific timestep. The core of our approach centers around Poly-Autoregressive models, which predict the future behavior of an agent during interaction by considering the agent's past state history and the state of other agents in the scene. In this paper, we develop the mechanics of Poly-Autoregressive (PAR) modeling and show that this framework applies without any modification to an extensive range of prediction problems that, on the surface, appear as entirely different scenarios, such as human action prediction in social situations, trajectory prediction for autonomous vehicles, and object pose prediction during hand-object interaction.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2735