Motivation & Description
Problem & Motivation Humanoid robots are dynamically stable systems. Existing safety control systems for industrial robots—which typically rely on suddenly bringing joints to a stop using brakes—do not work in this context. The robot would become a standing pendulum, fall over uncontrollably, and thus potentially cause greater danger.
The Project Solution The goal of the project is to develop an innovative, AI-based safety system for humanoid robots that enables functionally safe movement and behavior appropriate to the situation in everyday scenarios. To avoid relying on a cloud connection, all processes (such as safe human detection and behavior control) are fully integrated into a single embedded system.
- Change of Perspective: The system is trained so that it perceives the current situation and assesses hazardous situations directly from the robot’s subjective viewpoint (first-person perspective).
- AI Training with VLAs: Use of Vision-Language-Action Models (VLAs), which are trained using large datasets and generative AI to directly derive situationally safe robot actions from text, image, and video inputs.
- Adaptive Control Architecture: Systematic evaluation of sensor data quality and environmental context. Perceptual uncertainties are directly factored into control planning, enabling the robot to dynamically adapt its behavior.
- Novel Certification Process: Development of an industry-specific certification procedure as well as modular “Safety Assurance Cases” for the functional safety of humanoid robots working directly alongside humans.
Transfer and Utilization of Results
By the time the project is completed, it aims to achieve the following key technological results:
- All-in-One Hardware Controller: Development of a compact, energy-efficient, and AI-capable safety computer (certifiable to SIL 3, PLe) that serves as the central processing unit.
- Safe Human Detection: Real-time 360° human detection (SIL 2, PLd) for the safe localization and tracking of people and dynamic obstacles.
- Multiaxial Motion Control: Support and safe control of complex humanoid systems with over 50 degrees of freedom.
- Innovative Certification Process: The demonstration of a structured safety argumentation (“Safety Assurance Case”) that proves the robot meets all regulatory health and safety requirements in everyday use.
Synapticon
Specialist in motion control and drive technology. Develops the AI-based method for safely controlling the robot's behavior and motion.
