Analysis of Deep-Learning Methods in an ISO/TS 15066-Compliant Human-Robot Safety Framework

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Robot Interaction

Background:

• Collaborative robots (cobots) are increasingly used in manufacturing, requiring close human-robot proximity.
• Current safety standards (ISO/TS-15066) impose speed restrictions that can hinder collaborative task efficiency.
• Existing safety systems often lack the granularity to differentiate human body parts, leading to conservative operation.

Purpose of the Study:

• To introduce a novel deep-learning-based human-robot-safety framework (HRSF).
• To enable dynamic adaptation of robot velocities based on real-time human-robot separation distance.
• To maintain safety by adhering to biomechanical force and pressure limits while optimizing task efficiency.

Main Methods:

• Investigated four deep learning approaches for human body extraction: recognition, segmentation, pose estimation, and body part segmentation.
• Developed a framework that differentiates individual human body parts from the environment.
• Implemented dynamic velocity adjustments for robot control based on proximity and safety constraints.

Main Results:

• The proposed HRSF successfully differentiated human body parts, enabling more nuanced safety responses.
• Experiments showed a quantitative reduction in cycle time by up to 15% compared to conventional safety technologies.
• The framework demonstrated the potential for significantly improved efficiency in collaborative manufacturing tasks.

Conclusions:

• The deep-learning-based HRSF offers a significant advancement in human-robot collaboration safety and efficiency.
• By dynamically adapting robot behavior to individual human body parts, the framework overcomes limitations of conventional safety systems.
• This approach paves the way for more integrated and productive human-robot workspaces.