German startup Nucleus Robotics has emerged from stealth with an approach that puts humanoid robots into factories before they are fully autonomous. The company says it has already completed its first deployment at an undisclosed German industrial customer, with the implementation taking less than 90 days.
Rather than focusing primarily on developing another humanoid hardware platform, Nucleus wants to deploy and operate robots on behalf of industrial customers. Founder Melvin Schwarz describes the product as physical labor that customers pay for by the hour. Nucleus handles the deployment and operation of the robots and aims to gradually increase their autonomy using data collected on the factory floor.
The company argues that waiting for humanoids to become completely autonomous could slow industrial adoption. Instead, Nucleus combines AI with human supervision during the early stages of deployment. This allows robots to perform useful work while generating the real-world data needed to improve the underlying models.
From simulation to the factory floor
Nucleus uses a multi-stage deployment process. It first creates a 3D scan of the customer’s workplace. The environment is then recreated in simulation to identify potential problems and determine which robot configuration is suitable. The robots are subsequently installed at the customer site and initially operated under supervision.
For its first deployment, Nucleus says the robots received intensive supervision during the first three weeks. Data collected during this period is then used to enable the robots to handle more situations autonomously.
The first customer has not been named. Nucleus describes it as a German industrial company. Footage released by the startup shows a robot working in a logistics environment, handling boxes and containers. The robot is referred to by the company as Nucleus One.
The deployment also showed that standard robotic grippers are not suitable for every industrial task. For one handling process, in which workers largely grip boxes with their fingertips, Nucleus developed a dedicated end effector.
Real-world data to increase autonomy
Data from operational factory environments plays a central role in the company’s strategy. Nucleus says its software covers areas including perception, task planning, adaptation to the environment and the coordination of multiple robots. Each deployment is intended to generate additional examples of exceptions and unexpected situations that are difficult to anticipate in simulation.
Nucleus says its team includes people with previous experience at organizations and robotics companies including 1X, Foundation, Agile Robots, CERN, ESA and the United Nations.
Many details about the first deployment have yet to be disclosed. Nucleus has not revealed how many robots are involved, what productivity levels have been achieved, how often human intervention is required or what proportion of the work is currently performed fully autonomously. The company has also not disclosed which underlying humanoid platform is used for Nucleus One.
Schwarz says Nucleus aims to operate one of the larger deployed humanoid fleets worldwide by the end of 2027. Whether the model can scale will depend in part on how quickly human supervision can be reduced while maintaining reliable and productive factory operations.
Photo: Nucleus
