Industry Monitor Humanoid Industrial & Cobot AGV / AMR Quadruped Reducers · Servos · Sensors Drones & Autonomy Embodied AI
Robos News
Robotics

Arnold: A multi-task, multi-embodiment muscle transformer policy

arXiv:2508.18066v2 Announce Type: replace Abstract: Controlling high-dimensional and nonlinear musculoskeletal models of the human body is a foundational scientific challenge. Recent machine learning breakthroughs have heralded in-silico policies that master individual skills like reaching, object manipulation and locomotion in musculoskeletal systems with many degrees of freedom. However, these agents are merely "specialists", achieving high performance for a single skill. In this work, we dev

Published August 6, 2026 · Category: Robotics

Overview

arXiv:2508.18066v2 Announce Type: replace Abstract: Controlling high-dimensional and nonlinear musculoskeletal models of the human body is a foundational scientific challenge. Recent machine learning breakthroughs have heralded in-silico policies that master individual skills like reaching, object manipulation and locomotion in musculoskeletal systems with many degrees of freedom. However, these agents are merely "specialists", achieving high performance for a single skill. In this work, we develop Arnold, a transformer-based musculoskeletal control policy that masters multiple tasks and embodiments. Arnold combines behavior cloning and reinforcement learning to address 14 challenging control tasks spanning dexterous object manipulation, reaching, and locomotion, matching or exceeding the performance of single-task specialist policies. A key innovation is Arnold's sensorimotor vocabulary, a compositional representation of the semantics of heterogeneous sensory modalities, objectives, and actuators. Arnold leverages this vocabulary via a transformer architecture to deal with the variable observation and action spaces across tasks. This framework supports efficient multi-task, multi-embodiment learning and facilitates rapid adaptation to novel tasks, while encouraging universal motor strategies such as action and kinematic smoothness. Finally, causal probing of the motor output reveals that low-dimensional muscle synergies remain largely task-specific and that variance-based analyses systematically underestimate functional control dimensionality, consistent with biological observations on the limited transferability of such synergies. Code and data are available here: https://github.com/amathislab/arnold

Source

Originally published at arxiv.org.

Related Articles

Robos News Newsroom

Robos News reports on robotics research, components, manufacturers, field deployments, and industrial automation worldwide. Tip our newsroom: [email protected]

Email the newsroom →
Reporting standard: Product specifications, deployment counts, and performance claims are attributed to their source. Safety-critical decisions should be based on the applicable technical documentation and validation for the operating environment.
More from News →