Safeworld emerged publicly with a plan to test robots’ encounters with people by running their actual control software in simulated environments. The scenarios involve blind corners, falls and changing human appearances, with solar-construction company Gritt Robotics among its collaborators. The tests examine detection and stopping distance; the startup has not announced universal safety certification.
Safeworld plans simulated human encounters in which each robot runs the software that actually controls it. TechCrunch’s single report describes testing when a robot detects someone and whether its speed leaves enough distance to stop before contact. The startup emerged publicly on October 5. Ding Zhao, who directs the Safe AI Lab at Carnegie Mellon University, founded it with startup executive Kyle Wong and machine-learning engineer Simo Rachidi.[1]
Digital factory settings can include people carrying boxes, emerging around blind corners, stumbling or falling. Safeworld uses simulation infrastructure including Genesis and MuJoCo and intends to generate thousands of encounters with realistic human models. Zhao points to unstructured environments and differences between facilities’ safety standards as difficulties for assessment. The scenarios vary the conditions under which the same robot software must recognize a person and respond.[1]
Gritt Robotics, whose robots help install panels at industrial-scale solar farms, is collaborating with the startup. Its technology chief, Vishal Dugar, describes testing people kneeling or running, alongside variations in height, body shape, clothing and skin color. Safeworld wants to provide third-party safety evaluation for robot manufacturers. It is still deciding between a platform and a services-based offering; the report announces no independent certification proving safety across all robots.[1]
