analiza aiqbench
This arXiv paper argues that encouraging a model to predict actions from consecutive latent states helps its representations keep control-relevant distinctions that visual nuisance factors would otherwise wash out. The authors back this with a theoretical lower bound and experiments on Push-T, reporting planning success rising from 9.1% to 77.2% with their Concat IDM method. The results come from a single simulated benchmark across three training runs, so readers should treat the gains as promising evidence for this inductive bias rather than a demonstrated general improvement.
Learning Intrinsic Dynamics in Latent World Models
arXiv:2610.21612v1 Announce Type: new Abstract: Learning intrinsic dynamics---the relationship between actions and their consequences---is essential for world models. However, visual variations unrelated to the underlying control process can obscure this relationship in learned representations. We investigate the connection between action predictability, defined as the ability to reconstruct an action from consecutive latent states, and the preservation of intrinsic dynamics. Our central hypothesis is that encouraging action predictability discourages representations from discarding the distinctions between different actions and their consequences. We provide two complementary forms of evidence for this connection. Theoretically, we derive a positive lower bound on squared inverse prediction error when the encoder merges distinguishable outcomes of different actions from the same state and appearance, and show that forward prediction can still be exact on those transitions. Empirically, controlled appearance-diversity comparisons, supervision interventions, and latent probes support the role of action prediction in preserving control-relevant transition information. We instantiate action predictability with Concat IDM, which predicts actions from concatenated latent states and updates the shared encoder using action labels already available in offline trajectories, without changing the test-time planner. Across three independent training runs on Push-T with 24 appearances, Concat IDM increases planning success from 9.1% to 77.2%; its paired improvement is 52.9 percentage points larger than that observed with a single appearance. Together, these findings support action predictability as an inductive bias toward intrinsic dynamics and demonstrate the practical benefits of Concat IDM for latent planning under visual diversity.
Abstract:Learning intrinsic dynamics---the relationship between actions and their consequences---is essential for world models. However, visual variations unrelated to the underlying control process can obscure this relationship in learned representations. We investigate the connection between action predictability, defined as the ability to reconstruct an action from consecutive latent states, and the preservation of intrinsic dynamics. Our central hypothesis is that encouraging action predictability discourages representations from discarding the distinctions between different actions and their consequences. We provide two complementary forms of evidence for this connection. Theoretically, we derive a positive lower bound on squared inverse prediction error when the encoder merges distinguishable outcomes of different actions from the same state and appearance, and show that forward prediction can still be exact on those transitions. Empirically, controlled appearance-diversity comparisons, supervision interventions, and latent probes support the role of action prediction in preserving control-relevant transition information. We instantiate action predictability with Concat IDM, which predicts actions from concatenated latent states and updates the shared encoder using action labels already available in offline trajectories, without changing the test-time planner. Across three independent training runs on Push-T with 24 appearances, Concat IDM increases planning success from 9.1% to 77.2%; its paired improvement is 52.9 percentage points larger than that observed with a single appearance. Together, these findings support action predictability as an inductive bias toward intrinsic dynamics and demonstrate the practical benefits of Concat IDM for latent planning under visual diversity.
Artificial Intelligence (cs.AI)
arXiv:2610.21612 [cs.AI]
(or arXiv:2610.21612v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.21612 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 330 cuvinte
Subiecte conexe
DWM: Separating World Effects from Actions in Latent World Models
arXiv · 23 iul. 2026
DreamFormer: Dream Imitation with a Transformer World Model for Language-Conditioned Robotic Manipulation
arXiv · 6 oct. 2026
Learning PDE solution operators with variable initial conditions via Latent Dynamics Networks
arXiv · 7 oct. 2026
Interpreting learning dynamics of autoencoders: Transient scaling and emerging concepts of the Ising model
arXiv · 31 iul. 2026