Leveraging vast amounts of unlabeled internet video data for embodied AI is currently bottlenecked by the lack of action labels and the presence of action-correlated visual distractors. Although recent latent action policy optimization (LAPO) has shown promise in inferring proxy action labels from visual observations, its performance degrades significantly when distractors are present. To address this limitation, we propose a novel object-centric latent action learning framework that centers on objects rather than pixels. We leverage self-supervised object-centric pretraining to disentangle the movement of the agent and distracting background dynamics. This allows LAPO to focus on task-relevant interactions, resulting in more robust proxy-action labels, enabling better imitation learning and efficient adaptation of the agent with just a few action-labeled trajectories. We evaluated our method in eight visually complex tasks across the Distracting Control Suite (DCS) and Distracting MetaWorld (DMW). Our results show that object-centric pretraining mitigates the negative effects of distractors by 50%, as measured by downstream task performance: average return (DCS) and success rate (DMW).

Artem Shelmanov, Ekaterina Fadeeva, Akim Tsvigun, Ivan Tsvigun, Zhuohan Xie, Igor Kiselev, Nico Daheim, Caiqi Zhang, Artem Vazhentsev, Mrinmaya Sachan, Preslav Nakov, Timothy Baldwin. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Recently, latent action learning, pioneered by Latent Action Policies (LAPO), have shown remarkable pre-training efficiency on observation-only data, offering potential for leveraging vast amounts of video available on the web for embodied AI. However, prior work has focused on distractor-free data, where changes between observations are primarily explained by ground-truth actions. Unfortunately, real-world videos contain action-correlated distractors that may hinder latent action learning. Using Distracting Control Suite (DCS) we empirically investigate the effect of distractors on latent action learning and demonstrate that LAPO struggle in such scenario. We propose LAOM, a simple LAPO modification that improves the quality of latent actions by 8x, as measured by linear probing. Importantly, we show that providing supervision with ground-truth actions, as few as 2.5% of the full dataset, during latent action learning improves downstream performance by 4.2x on average. Our findings suggest that integrating supervision during Latent Action Models (LAM) training is critical in the presence of distractors, challenging the conventional pipeline of first learning LAM and only then decoding from latent to ground-truth actions.

Previously proposed variational techniques for approximate MMAP inference in complex graphical models of high-order factors relax a dual variational objective function to obtain its tractable approximation, and further perform MMAP inference in the resulting simplified graphical model, where the sub-graph with decision variables is assumed to be a disconnected forest. In contrast, we developed novel variational MMAP inference algorithms and proximal convergent solvers, where we can improve the approximation accuracy while better preserving the original MMAP query by designing such a dual variational objective function that an upper bound approximation is applied only to the entropy of decision variables. We evaluate the proposed algorithms on both simulated synthetic datasets and diagnostic Bayesian networks taken from the UAI inference challenge, and our solvers outperform other variational algorithms in a majority of reported cases. Additionally, we demonstrate the important real-life application of the proposed variational approaches to solve complex tasks of policy optimization by MMAP inference, and performance of the implemented approximation algorithms is compared. Here, we demonstrate that the original task of optimizing POMDP controllers can be approached by its reformulation as the equivalent problem of marginal-MAP inference in a novel single-DBN generative model, which guarantees that the control policies computed by probabilistic inference over this model are optimal in the traditional sense. Our motivation for approaching the planning problem through probabilistic inference in graphical models is explained by the fact that by transforming a Markovian planning problem into the task of probabilistic inference (a marginal MAP problem) and applying belief propagation techniques in generative models, we can achieve a computational complexity reduction from PSPACE-complete or NEXP-complete to NPPP-complete in comparison to solving the POMDP and Dec-POMDP models respectively search vs. dynamic programming).

As a promising alternative to using standard (often intractable) planning techniques with Bellman equations, we propose an interesting method of optimizing POMDP controllers by probabilistic inference in a novel equivalent single-DBN generative model. Our inference approach to POMDP planning allows for (1) for application of various techniques for probabilistic inference in single graphical models, and (2) for exploiting the factored structure in a controller architecture to take advantage of natural structural constrains of planning problems and represent them compactly. Our contributions can be summarized as follows: (1) we designed a novel single-DBN generative model that ensures that the task of probabilistic inference is equivalent to the original problem of optimizing POMDP controllers, and (2) we developed several inference approaches to approximate the value of the policy when exact inference methods are not tractable to solve large-size problems with complex graphical models. The proposed approaches to policy optimization by probabilistic inference are evaluated on several POMDP benchmark problems and the performance of the implemented approximation algorithms is compared.