aiqbench analysis
MoPA introduces a novel framework for mobile manipulation by aligning dual perceptual streams with kinematically coupled action modalities using a structured Mixture-of-Transformers decoder. This approach achieves state-of-the-art performance on benchmarks and significantly improves real-world task success rates by explicitly modeling subsystem-specific perception-action correspondence. The work advances the field of embodied AI by resolving the implicit nature of traditional perception-action mappings in complex robotic systems.
MoPA: Coordinated Mobile Manipulation via Subsystem-Specific Perception Alignment
arXiv:2609.12081v1 Announce Type: new Abstract: Mobile manipulation requires perceptual evidence at different spatial scales for base motion and arm control, while the two action modalities remain kinematically coupled. Existing policies often employ specialized action generation for different subsystems but condition heterogeneous action branches on a shared perceptual representation, leaving subsystem-specific perception-action correspondence implicit. We present MoPA, a framework that aligns perceptual conditioning with mobility and manipulation while preserving coordination at the action level. Dual Perceptual Streams employ two mutually masked query banks to extract separate perceptual representations from a shared vision-language context. Perception2Action Adaptation jointly updates each query bank and its corresponding action stream at every layer of a structured Mixture-of-Transformers decoder, while enabling information exchange between the two action streams. Coupled conditional flow matching learns a joint vector field for coordinated generation of both action chunks. On the ManiSkill-HAB benchmark, MoPA achieves state-of-the-art performance across all three task suites. Across four real-world tasks, MoPA achieves a mean full-task success rate of 76.3%, outperforming the best baseline by 12.5 percentage points. Ablation studies and further analyses validate the effectiveness of the proposed design. Website is available at: https://mopa-policy.github.io/.
Abstract:Mobile manipulation requires perceptual evidence at different spatial scales for base motion and arm control, while the two action modalities remain kinematically coupled. Existing policies often employ specialized action generation for different subsystems but condition heterogeneous action branches on a shared perceptual representation, leaving subsystem-specific perception-action correspondence implicit. We present MoPA, a framework that aligns perceptual conditioning with mobility and manipulation while preserving coordination at the action level. Dual Perceptual Streams employ two mutually masked query banks to extract separate perceptual representations from a shared vision-language context. Perception2Action Adaptation jointly updates each query bank and its corresponding action stream at every layer of a structured Mixture-of-Transformers decoder, while enabling information exchange between the two action streams. Coupled conditional flow matching learns a joint vector field for coordinated generation of both action chunks. On the ManiSkill-HAB benchmark, MoPA achieves state-of-the-art performance across all three task suites. Across four real-world tasks, MoPA achieves a mean full-task success rate of 76.3%, outperforming the best baseline by 12.5 percentage points. Ablation studies and further analyses validate the effectiveness of the proposed design. Website is available at: this https URL.
Robotics (cs.RO); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2609.12081 [cs.RO]
(or arXiv:2609.12081v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.12081 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.
Full text under the publisher's open licence · 284 words
Related coverage
Praxis: Distilling Physical Interaction Priors from Egocentric Videos for Generalizable Whole-Body Manipulation
arXiv · Sep 28, 2026
Game Design Creativity evals
LessWrong · Oct 12, 2026
NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI
NVIDIA · May 21, 2026
Building a Foundation Stack for General-Purpose Robots
IEEE Spectrum · Jul 13, 2026