Hannover Messe

Siemens tests autonomous humanoid AI robot in Erlangen plant

Siemens has deployed an HMND-01 robot for logistics tasks to test its capabilities for building an AI-driven manufacturing facility.

listen Print view
Robot places a container.

The HMND-01 robot placing a container.

(Image: Siemens AG)

3 min. read

Siemens, together with the British robotics company Humanoid, tested the HMND 01 humanoid robot in a Siemens plant in Erlangen. The wheeled, semi-humanoid robot performed logistics tasks autonomously in the completely AI-controlled plant. The use of Nvidia's AI components drastically reduced the development time of the robot for use in the Siemens plant.

At Siemens, the semi-humanoid robot HMND 01, consisting of a humanoid torso mounted on an omnidirectionally mobile platform, was deployed. The robot uses 360-degree cameras and depth sensors to recognize its environment. The robot can adjust its height up to 2.2 m. With its two arms, it can lift weights up to 15 kg. Higher weights can also be carried if they are closer to the body. The reach of the arms, when retracted, extends from the floor to a height of about 2 m. Combined with its height adjustment, this allows it to reach shelves and retrieve parts from shelves up to 60 cm deep.

In total, the robot has 29 degrees of freedom. A humanoid hand with twelve degrees of freedom or a two-finger gripper can be used with the HMND 01 Alpha, depending on the application.

For logistics tasks at Siemens, a two-finger gripper was used, with which the robot autonomously gripped, transported, and placed containers. According to Siemens, all required key performance indicators were achieved. The robot achieved around 60 container movements per hour during an operating time of eight hours. The pick-and-place success rate was over 90 percent.

For the development and optimization of the robot for logistics tasks at Siemens, Humanoid integrated Nvidia's complete Physical AI Stack into the HMND-01 robot. For edge computing, Humanoid uses Jetson Thor, for simulation Isaac Sim, and Isaac Lab for Reinforcement Learning (RL) and policy training. Through simulations, the robot could be precisely adapted to its tasks, for example, in the selection of actuators. The required joint force and the distribution of masses could be determined in advance, and the robot optimized accordingly. A prototype could thus be developed particularly quickly: Instead of the usual 18 to 24 months, development took only seven months.

Videos by heise

Humanoid's robot is to be fully integrated into a manufacturing system. This includes, for example, data exchange with production systems and autonomous transport systems. Furthermore, the robot must adapt to the workflows of other machines and people and be able to react to possible changes. Siemens intends to achieve this by using a digital twin of a production facility. Siemens Xcelerator combines different systems under one roof for this purpose. This makes it possible, for example, to integrate robots into production. Siemens' goal is to build a fully AI-controlled, adaptive manufacturing facility.

(olb)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.