MARS Policy: Multimodality Only When It Matters

• URL: http://arxiv.org/abs/2605.29766v1
• Date: Thu, 28 May 2026 11:12:30 GMT
• Title: MARS Policy: Multimodality Only When It Matters
• Abstract summary: multimodality, which enables robots to capture diverse yet valid behavioral patterns, has driven the rapid emergence of generative policies as a dominant paradigm in robot learning.<n>We propose the Modality-Adaptive Robot Sampling (MARS) policy, which adaptively invokes tailored multimodality only when it is truly beneficial.<n>MARS bridges the gap between the multimodal capability of generative policies and the superior training and inference efficiency of deterministic models.
• Score: 26.027481334743854
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Imitation learning has become a cornerstone for solving complex robotic manipulation tasks. In particular, multimodality, which enables robots to capture diverse yet valid behavioral patterns, has driven the rapid emergence of generative policies as a dominant paradigm in robot learning. However, achieving such multimodality typically relies on stochastic noise initialization and iterative denoising procedures, resulting in substantial training complexity and low inference efficiency. Meanwhile, not all phases of a robotic task inherently require behavioral diversity. Motivated by this insight, we propose the Modality-Adaptive Robot Sampling (MARS) policy, which adaptively invokes tailored stochasticity only when it is truly beneficial, while reverting to an efficient deterministic learning during single-modal phases. In other words, the proper amount of noise is injected only at the proper time. By selectively activating multimodal generation, MARS policy bridges the gap between the multimodal capability of generative policies and the superior training and inference efficiency of deterministic models. Empirical studies across 8 simulated and 4 real-world tasks demonstrate that MARS exhibits robust multimodal expressivity and high efficiency, with a 16.67% success rate improvement and an 83.20% inference latency reduction in real-world tests. Counterintuitively, MARS also outpaces deterministic policies in training efficiency on near-deterministic tasks by more effectively modeling nuanced action diversity.

Related papers

• Scaling Tasks, Not Samples: Mastering Humanoid Control through Multi-Task Model-Based Reinforcement Learning [49.82882141491629]
  We argue that effective online learning should scale the emphnumber of tasks, rather than the number of samples per task.<n>This regime reveals a structural advantage of model-based reinforcement learning.<n>We instantiate this idea with textbfEfficientZero-Multitask (EZ-M), a sample-efficient multi-task algorithm for online learning.
  arXiv  Detail & Related papers  (2026-03-02T05:07:43Z)
• Abstracting Robot Manipulation Skills via Mixture-of-Experts Diffusion Policies [13.826293542795938]
  Diffusion-based policies have recently shown strong results in robot manipulation, but their extension to multi-task scenarios is hindered by the high cost of scaling model size and demonstrations.<n>We introduce Skill Mixture-of-Experts Policy (SMP), a diffusion-based mixture-of-experts policy that learns a compact skill basis and uses sticky routing to compose actions from a small, task-relevant subset of experts at each step.<n>We validate SMP in simulation and on a real dual-arm platform with multi-task learning and transfer learning tasks, where SMP achieves higher success rates and markedly lower inference cost than large diffusion baselines.
  arXiv  Detail & Related papers  (2026-01-29T04:17:56Z)
• From Sparse Decisions to Dense Reasoning: A Multi-attribute Trajectory Paradigm for Multimodal Moderation [59.27094165576015]
  We propose a novel learning paradigm (UniMod) that transitions from sparse decision-making to dense reasoning traces.<n>By constructing structured trajectories encompassing evidence grounding, modality assessment, risk mapping, policy decision, and response generation, we reformulate monolithic decision tasks into a multi-dimensional boundary learning process.<n>We introduce specialized optimization strategies to decouple task-specific parameters and rebalance training dynamics, effectively resolving interference between diverse objectives in multi-task learning.
  arXiv  Detail & Related papers  (2026-01-28T09:29:40Z)
• Time-Unified Diffusion Policy with Action Discrimination for Robotic Manipulation [19.449168375853347]
  We present Time-Unified Diffusion Policy (TUDP), which utilizes action recognition capabilities to build a time-unified denoising process.<n>Our method achieves state-of-the-art performance on RLBench with the highest success rate of 82.6% on a multi-view setup and 83.8% on a single-view setup.
  arXiv  Detail & Related papers  (2025-06-11T06:11:49Z)
• Unified Generative and Discriminative Training for Multi-modal Large Language Models [88.84491005030316]
  Generative training has enabled Vision-Language Models (VLMs) to tackle various complex tasks. Discriminative training, exemplified by models like CLIP, excels in zero-shot image-text classification and retrieval. This paper proposes a unified approach that integrates the strengths of both paradigms.
  arXiv  Detail & Related papers  (2024-11-01T01:51:31Z)
• Guided Reinforcement Learning for Robust Multi-Contact Loco-Manipulation [12.377289165111028]
  Reinforcement learning (RL) often necessitates a meticulous Markov Decision Process (MDP) design tailored to each task. This work proposes a systematic approach to behavior synthesis and control for multi-contact loco-manipulation tasks. We define a task-independent MDP to train RL policies using only a single demonstration per task generated from a model-based trajectory.
  arXiv  Detail & Related papers  (2024-10-17T17:46:27Z)
• Active Fine-Tuning of Multi-Task Policies [54.65568433408307]
  We propose AMF (Active Multi-task Fine-tuning) to maximize multi-task policy performance under a limited demonstration budget.<n>We derive performance guarantees for AMF under regularity assumptions and demonstrate its empirical effectiveness in complex and high-dimensional environments.
  arXiv  Detail & Related papers  (2024-10-07T13:26:36Z)
• Sparse Diffusion Policy: A Sparse, Reusable, and Flexible Policy for Robot Learning [61.294110816231886]
  We introduce a sparse, reusable, and flexible policy, Sparse Diffusion Policy (SDP) SDP selectively activates experts and skills, enabling efficient and task-specific learning without retraining the entire model. Demos and codes can be found in https://forrest-110.io/sparse_diffusion_policy/.
  arXiv  Detail & Related papers  (2024-07-01T17:59:56Z)
• Multimodal Instruction Tuning with Conditional Mixture of LoRA [51.58020580970644]
  This paper introduces a novel approach that integrates multimodal instruction tuning with Low-Rank Adaption (LoRA)<n>It innovates upon LoRA by dynamically constructing low-rank adaptation matrices tailored to the unique demands of each input instance.<n> Experimental results on various multimodal evaluation datasets indicate that MixLoRA not only outperforms the conventional LoRA with the same or even higher ranks.
  arXiv  Detail & Related papers  (2024-02-24T20:15:31Z)
• Asynchronous Multi-Agent Reinforcement Learning for Efficient Real-Time Multi-Robot Cooperative Exploration [16.681164058779146]
  We consider the problem of cooperative exploration where multiple robots need to cooperatively explore an unknown region as fast as possible. Existing MARL-based methods adopt action-making steps as the metric for exploration efficiency by assuming all the agents are acting in a fully synchronous manner. We propose an asynchronous MARL solution, Asynchronous Coordination Explorer (ACE), to tackle this real-world challenge.
  arXiv  Detail & Related papers  (2023-01-09T14:53:38Z)
• Strategy Discovery and Mixture in Lifelong Learning from Heterogeneous Demonstration [1.2891210250935146]
  Learning from Demonstration (LfD) approaches empower end-users to teach robots novel tasks via demonstrations of the desired behaviors. In this paper, we propose a novel algorithm, Dynamic Multi-Strategy Reward Distillation (DMSRD), which distills common knowledge between heterogeneous demonstrations. Our personalized, federated, and lifelong LfD architecture surpasses benchmarks in two continuous control problems with an average 77% improvement in policy returns and 42% improvement in log likelihood.
  arXiv  Detail & Related papers  (2022-02-14T20:10:25Z)

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.