A Mixed-Perception Approach for Safe Human-Robot Collaboration in Industrial Automation

Area of Science:

• Robotics
• Artificial Intelligence
• Manufacturing Systems

Background:

• Digital manufacturing demands high automation but faces challenges with human presence in shared workspaces.
• Robots lack awareness of human position and intent, creating safety concerns and reducing productivity.
• Existing systems struggle to balance automation efficiency with human-robot collaboration (HRC) safety.

Purpose of the Study:

• To design a reliable safety monitoring system for collaborative robots (cobots).
• To enhance human safety in shared workspaces by integrating visual and tactile perception.
• To improve cobot awareness of human intentions for more productive HRC.

Main Methods:

• Developed a safety monitoring system combining visual human action recognition and tactile human-robot contact interpretation.
• Collected datasets using volunteers, capturing both visual (skeleton representation) and contact data.
• Utilized two distinct deep learning networks for action recognition and contact detection.

Main Results:

• The integrated system demonstrated effective recognition of human actions and differentiation between intentional/incidental physical contact.
• Achieved promising results in enhancing safety and increasing cobot perception of human intentions.
• Validated the approach using datasets from multiple volunteers.

Conclusions:

• The developed system offers a reliable approach to improving safety in human-robot collaboration.
• This AI-driven solution paves the way for safer and more productive industrial automation.
• Future work can further refine perception and adaptiveness for advanced HRC.