Google DeepMind humanoids test automotive kitting at Mercedes-Benz factories

Google DeepMind has announced a new proof of concept involving the deployment of Apptronik’s Apollo 2 humanoids on active Mercedes-Benz production lines. The integration of DeepMind AI technology aims to streamline the preparation of vehicle assembly by automating the organization of parts, a process known as automotive kitting.

Testing DeepMind AI in manufacturing

According to the disclosure, the robots achieved a 96% success rate while operating at Mercedes-Benz facilities in Kecskemét, Hungary, and Sindelfingen, Germany. The primary task involved picking metal brackets from bulk bins and arranging them into specific sequencing containers. This process is critical for ensuring that assembly line workers have the exact components required for each vehicle in the correct order.

To manage these tasks, the system utilizes two distinct models. The Gemini Robotics ER model is responsible for tracking the overall workflow, while the Gemini Robotics 2.5 model provides the motor control for the robot's arms and hands. Apptronik CEO Jeff Cardenas noted that the models showed significant improvement following a few weeks of post-training and were able to transition to the factory floor with minimal performance degradation.

Apollo 2 humanoids achieved a 96% success rate sorting automotive parts on active production lines.

This collaboration builds upon an agreement established in March 2024, which sought to explore how Apollo humanoids could assist with assembly kits, component inspection, and container transport. The long-term goal is to utilize humanoids to perform physically demanding and repetitive tasks in environments originally designed for human workers, potentially eliminating the need for expensive factory redesigns.

While the results are being presented as a successful test of autonomous kitting, DeepMind has not provided specific details regarding the total trial count, operating costs, or the frequency of human intervention required to maintain production. The company also did not define the exact metrics used to calculate the success rate, such as whether it refers to individual picks or completed trays.

Why it matters

The deployment of humanoid robots in automotive manufacturing represents a shift toward automating complex, human-centric tasks on the assembly line. For factory operators, the key challenge remains determining the efficiency of these systems during a full shift and the extent of human oversight needed to prevent production bottlenecks. As companies continue to integrate advanced robotics, the ability to maintain consistent uptime without frequent manual resets will be a deciding factor in the widespread adoption of this technology.