# Learning Additively Compositional Latent Actions for Embodied AI

Source: [Learning Additively Compositional Latent Actions for Embodied AI](<https://www.sota2.com/research/paper/69d48f9a7d0172140ec74363>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-08T05:00:04.658Z

Authors: Hangxing Wei, Xiaoyu Chen, Chuheng Zhang, Tim Pearce, Jianyu Chen, Alex Lamb, Li Zhao, Jiang Bian

Published: 2026-04-03T08:53:29.000Z

## Abstract

Latent action learning infers pseudo-action labels from visual transitions, providing an approach to leverage internet-scale video for embodied AI. However, most methods learn latent actions without structural priors that encode the additive, compositional structure of physical motion. As a result, latents often entangle irrelevant scene details or information about future observations with true state changes and miscalibrate motion magnitude. We introduce Additively Compositional Latent Action Model (AC-LAM), which enforces scene-wise additive composition structure over short horizons on the latent action space. These AC constraints encourage simple algebraic structure in the latent action space~(identity, inverse, cycle consistency) and suppress information that does not compose additively. Empirically, AC-LAM learns more structured, motion-specific, and displacement-calibrated latent actions and provides stronger supervision for downstream policy learning, outperforming state-of-the-art LAMs across simulated and real-world tabletop tasks.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Tabletop Manipulation Policy Learning on Emoji Table-Top GrinningFace (ID)](<https://www.sota2.com/research/sota/tabletop-manipulation-policy-learning-on-emoji-table-top-grinningface-id.md>): Tabletop Manipulation Policy Learning; Emoji Table-Top GrinningFace (ID); Success (S): 55.

- [Tabletop Manipulation on Real-World Tabletop Manipulation (In-Distribution)](<https://www.sota2.com/research/sota/tabletop-manipulation-on-real-world-tabletop-manipulation-in-distribution.md>): Tabletop manipulation; Real-World Tabletop Manipulation (In-Distribution); Success Rate: 60.

- [Tabletop Manipulation on Real-World Tabletop Manipulation Out-of-Distribution Distractors](<https://www.sota2.com/research/sota/tabletop-manipulation-on-real-world-tabletop-manipulation-out-of-distribution-distractors.md>): Tabletop manipulation; Real-World Tabletop Manipulation Out-of-Distribution Distractors; Success Rate: 53.3.

- [Tabletop Manipulation on Real-World Tabletop Manipulation OOD-B (Out-of-Distribution Backgrounds)](<https://www.sota2.com/research/sota/tabletop-manipulation-on-real-world-tabletop-manipulation-ood-b-out-of-distribution-backgrounds.md>): Tabletop manipulation; Real-World Tabletop Manipulation OOD-B (Out-of-Distribution Backgrounds); Success Rate: 33.3.

- [Tabletop Manipulation Policy Learning on Emoji Table-Top GrinningFace (train)](<https://www.sota2.com/research/sota/tabletop-manipulation-policy-learning-on-emoji-table-top-grinningface-train.md>): Tabletop Manipulation Policy Learning; Emoji Table-Top GrinningFace (train); Success Rate (S): 42.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.03340>)

[PDF](<https://arxiv.org/pdf/2604.03340.pdf>)