Emergent Compositional Skills in Mixture-of-Experts VLAs

• URL: http://arxiv.org/abs/2607.20771v2
• Date: Fri, 24 Jul 2026 19:25:04 GMT
• Title: Emergent Compositional Skills in Mixture-of-Experts VLAs
• Abstract summary: We ask whether a VLA trained with a simplified Mixture-of-Experts (MoE) action head can emergently learn to decompose tasks into reusable, interpretable primitives.<n>We find that learned experts are heavily reused across tasks and consistently correspond to qualitatively distinct low-level behaviors.
• Score: 4.404201871360367
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We consider the problem of learning compositional robot policies end-to-end from expert demonstrations, without any pre-specified notion of task decomposition or hierarchy. We ask whether a VLA trained with a simplified Mixture-of-Experts (MoE) action head can emergently learn to decompose tasks into reusable, interpretable primitives. We find that learned experts are heavily reused across tasks and consistently correspond to qualitatively distinct low-level behaviors, suggesting that the router implicitly learns to perform high-level sequencing while experts serve as compositional primitives. Our MoE matches the task performance of a monolithic baseline while demonstrating meaningful expert specialization, a step toward modular, interpretable robot policies that emerge from data alone.

Related papers

• Sparse Subspace-to-Expert Sharing for Task-Agnostic Continual Learning [10.01449025634975]
  Existing methods treat parameters uniformly, failing to distinguish between specific task knowledge and shared capabilities.<n>We introduce Mixture of Sparse Experts for Task Agnostic Continual Learning (SETA), a framework that resolves the plasticity-stability conflict through adaptive sparse subspace decomposition.<n>We show SETA achieves competitive or superior overall performance relative to state-of-the-art continual learning baselines.
  arXiv  Detail & Related papers  (2026-06-05T17:53:52Z)
• LAR-MoE: Latent-Aligned Routing for Mixture of Experts in Robotic Imitation Learning [1.7272266547709354]
  We introduce Latent-Aligned Routing for Mixture of Experts (LAR-MoE), a two-stage framework that decouples unsupervised skill discovery from policy learning.<n>In pre-training, we learn a joint latent representation between observations and future actions through student-teacher co-training.<n>In a post-training stage, the expert routing is regularized to follow the structure of the learned latent space, preventing expert collapse.
  arXiv  Detail & Related papers  (2026-03-09T15:10:31Z)
• Phase-Aware Mixture of Experts for Agentic Reinforcement Learning [23.18318273534301]
  A plausible remedy could be employing the Mixture-of-Experts (MoE) architecture in the policy network.<n>MoE allows different parameters (experts) to specialize in different tasks, preventing simple tasks from dominating all parameters.<n>We propose textbfPhase-Aware Mixture of Experts (PA-MoE).<n>It first features a lightweight emphphase router that learns latent phase boundaries directly from the RL objective without pre-defining phase categories.<n>Then, the phase router allocates temporally consistent assignments to the same expert, allowing experts to preserve phase-specific expertise
  arXiv  Detail & Related papers  (2026-02-19T03:18:30Z)
• SAME: Stabilized Mixture-of-Experts for Multimodal Continual Instruction Tuning [83.66308307152808]
  We propose StAbilized Mixture-of-Experts (SAME) for Multimodal Continual Instruction Tuning (MCIT)<n>SAME stabilizes expert selection by decomposing routing dynamics into subspaces and updating only task-relevant directions.<n>It also introduces adaptive expert activation to freeze selected experts during training, reducing redundant and cross-task interference.
  arXiv  Detail & Related papers  (2026-02-02T11:47:06Z)
• Split-on-Share: Mixture of Sparse Experts for Task-Agnostic Continual Learning [10.01449025634975]
  Continual learning in Large Language Models (LLMs) is hindered by the plasticity-stability dilemma.<n>We introduce SETA, a framework that resolves the plasticity-stability conflict by decomposing the model into modular subspaces.<n>We show that SETA consistently outperforms state-of-the-art parameter-efficient fine-tuning-based continual learning methods.
  arXiv  Detail & Related papers  (2026-01-24T22:39:22Z)
• One-Prompt Strikes Back: Sparse Mixture of Experts for Prompt-based Continual Learning [52.966712416640085]
  We propose SMoPE, a novel framework that integrates the benefits of both task-specific and shared prompt strategies.<n>SMoPE consistently outperforms task-specific prompt methods and achieves performance competitive with state-of-the-art approaches.
  arXiv  Detail & Related papers  (2025-09-29T08:54:58Z)
• Complexity Experts are Task-Discriminative Learners for Any Image Restoration [80.46313715427928]
  We introduce complexity experts" -- flexible expert blocks with varying computational complexity and receptive fields.<n>This preference effectively drives task-specific allocation, assigning tasks to experts with the appropriate complexity.<n>The proposed MoCE-IR model outperforms state-of-the-art methods, affirming its efficiency and practical applicability.
  arXiv  Detail & Related papers  (2024-11-27T15:58:07Z)
• Pre-training Multi-task Contrastive Learning Models for Scientific Literature Understanding [52.723297744257536]
  Pre-trained language models (LMs) have shown effectiveness in scientific literature understanding tasks. We propose a multi-task contrastive learning framework, SciMult, to facilitate common knowledge sharing across different literature understanding tasks.
  arXiv  Detail & Related papers  (2023-05-23T16:47:22Z)
• Inferring Versatile Behavior from Demonstrations by Matching Geometric Descriptors [72.62423312645953]
  Humans intuitively solve tasks in versatile ways, varying their behavior in terms of trajectory-based planning and for individual steps. Current Imitation Learning algorithms often only consider unimodal expert demonstrations and act in a state-action-based setting. Instead, we combine a mixture of movement primitives with a distribution matching objective to learn versatile behaviors that match the expert's behavior and versatility.
  arXiv  Detail & Related papers  (2022-10-17T16:42:59Z)
• EEML: Ensemble Embedded Meta-learning [5.9514420658483935]
  We propose an ensemble embedded meta-learning algorithm (EEML) that explicitly utilizes multi-model-ensemble to organize prior knowledge into diverse specific experts. We rely on a task embedding cluster mechanism to deliver diverse tasks to matching experts in training process and instruct how experts collaborate in test phase. The experimental results show that the proposed method outperforms recent state-of-the-arts easily in few-shot learning problem.
  arXiv  Detail & Related papers  (2022-06-18T12:37:17Z)
• Learning Task Decomposition with Ordered Memory Policy Network [73.3813423684999]
  We propose Ordered Memory Policy Network (OMPN) to discover subtask hierarchy by learning from demonstration. OMPN can be applied to partially observable environments and still achieve higher task decomposition performance. Our visualization confirms that the subtask hierarchy can emerge in our model.
  arXiv  Detail & Related papers  (2021-03-19T18:13:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.