ARMAR-6: A High-Performance Humanoid for Human-Robot Collaboration in Real-World Scenarios

Tamim Asfour, Fabian Paus, Mirko Wächter, Lukas Kaul, Samuel Rader, Pascal Weiner, Simon Ottenhaus, Raphael Grimm, You Zhou, Markus GrotzIEEE Robotics Autom. Mag 2019 · OpenReview

Abstract

A major goal of humanoid robotics is to enable safe and reliable human-robot collaboration in real-world scenarios. In this article, we present ARMAR-6, a new high-performance humanoid robot for various tasks, including but not limited to grasping, mobile manipulation, integrated perception, bimanual collaboration, compliant-motion execution, and natural language understanding. We describe how the requirements arising from these tasks influenced our major design decisions, resulting in vertical integration during the joint hardware and software development phases. In particular, the entire hardware—including its structure, sensor-actuator units, and low-level controllers—as well as its perception, grasping and manipulation skills, task coordination, and the entire software architecture were all developed by one team of engineers. Component interaction is facilitated by our software framework ArmarX, which further facilitates the seamless integration and interchange of third-party contributions. To showcase the robot's capabilities, we present its performance in a challenging industrial maintenance scenario that requires human-robot collaboration, where the robot autonomously recognizes the human's need of help and offers said help in a proactive way.

Introduction

Humanoid robots have long been confined to research laboratories because the complexities of real-world collaboration are immense. To be truly useful in industrial settings like factories or automated warehouses, a robot must transcend simple reactive behaviors. It must be physically capable of handling heavy tools, safe enough to work alongside humans without physical barriers, and intelligent enough to anticipate a coworker's needs.

ARMAR-6 collaborating with a human technician.

Developed by the Karlsruhe Institute of Technology (KIT) as part of the European SecondHands project, ARMAR-6 is designed to act as a proactive assistant for human technicians. The project addresses the "vertical integration" challenge—designing everything from motor controllers to natural language processors—to ensure these systems work together seamlessly in real-time.

Hardware-Software Co-design

The development of ARMAR-6 utilized a rigorous hardware-software co-design approach to meet the demands of industrial maintenance.

Modular Hardware

The robot features custom Sensor-Actuator-Controller (SAC) units. These self-contained joints integrate motors, gears, and electronics into single modules, allowing for easier maintenance and scalability. A notable design feature is the use of slip rings for cabling in the shoulder, upper arm, and forearm, providing "limitless" rotation that prevents the robot from becoming tangled during complex bimanual manipulations.

Frontal view of ARMAR-6 showcasing its bimanual manipulation capabilities.

ArmarX Software Framework

The software architecture is built on the ArmarX middleware, which organizes the robot’s cognitive functions into three distinct levels:

  • Low Level: 1 kHz real-time control for torque and velocity.
  • Middle Level: Perception and memory (MemoryX), handling object localization and marker-free 3D human pose estimation.
  • High Level: Task coordination using statecharts and symbolic planning.

Proactive Assistance and Intelligence

A core innovation of ARMAR-6 is its Recognition of the Need of Help (RoNoH). By combining visual data (observing a human's posture) and haptic feedback (sensing weight shifts), the robot can deduce when a technician is struggling with a heavy object and step in to assist without being explicitly commanded.

To support these interactions, the robot utilizes several state-of-the-art data-driven models:

  • Perception: OpenPose and "Lifting from the Deep" are used for human pose estimation.
  • Grasping: Deep learning neural networks generate grasp candidates for both familiar and unknown objects.
  • Communication: An attention-based encoder-decoder network provides end-to-end speech-to-intent processing for natural language understanding (NLU).

Experimental Results and Validation

ARMAR-6 stands 192 cm tall and weighs 160 kg, boasting a significant 10 kg payload capacity per arm at a full 1.3 m reach. Powered by a 1.8 kWh battery, it is capable of approximately four hours of autonomous operation.

The system was validated in a real-world automated fulfillment center in the UK. During these trials, the robot successfully demonstrated bimanual collaboration, such as carrying a heavy guard panel with a human and automatically adjusting its grip when it sensed an unbalanced load. In user studies, technicians evaluated the robot using the Godspeed Questionnaire, generally perceiving ARMAR-6 as a helpful, safe, and reliable coworker.

Conclusion and Future Outlook

While ARMAR-6 represents a significant leap in humanoid capability, the authors identify several areas for future development. Current research is focused on improving failure recovery, allowing the robot to detect when a task is failing and autonomously attempt a new strategy. Additionally, while the robot can grasp unknown objects, its "dexterous manipulation" of small tools requires further refinement. Finally, the team is developing an "episodic memory" system to allow the robot to remember past interactions with specific technicians, enabling it to better predict individual needs over time.

Summary updated

Abstract

A major goal of humanoid robotics is to enable safe and reliable human-robot collaboration in realworld scenarios. In this article, we present ARMAR-6, a new high-performance humanoid robot for various tasks, including but not limited to grasping, mobile manipulation, integrated perception, bimanual collaboration, compliant-motion execution, and natural language understanding. We describe how the requirements arising from these tasks influenced our major design decisions, resulting in vertical integration during the joint hardware and software development phases. In particular, the entire hardware-including its structure, sensor-actuator units, and low-level controllers-as well as its perception, grasping and manipulation skills, task coordination, and the entire software architecture were all developed by one team of engineers. Component interaction is facilitated by our software framework ArmarX, which further facilitates the seamless integration and interchange of third-party contributions. To showcase the robot's capabilities, we present its performance in a challenging industrial maintenance scenario that requires human-robot collaboration, where the robot autonomously recognizes the human's need of help and offers said help in a proactive way.
BibTeX
@inproceedings{asfour2019armar,
  author = {Tamim Asfour and Fabian Paus and Mirko Wächter and Lukas Kaul and Samuel Rader and Pascal Weiner and Simon Ottenhaus and Raphael Grimm and You Zhou and Markus Grotz},
  title = {{ARMAR-6: A High-Performance Humanoid for Human-Robot Collaboration in Real-World Scenarios}},
  booktitle = {IEEE Robotics Autom. Mag},
  year = {2019},
  url = {https://openreview.net/forum?id=qIwZXx-Mv9},
}
Figure