Safeworld raises more than $12 million to assess robot safety

Safeworld has raised more than $12 million in seed funding to assess robotic control systems in simulations with realistic human models. Shine Capital and a16z Speedrun led the round as the company emerged from stealth. Its stated focus is testing how robots behave around people before deployment in settings where conditions vary.

The funding announcement places Safeworld in the AI and robotics safety ecosystem. The company was founded by Dr. Ding Zhao, Kyle Wong and Simo Rachidi. Zhao directs the Safe AI lab at Carnegie Mellon University, while the source describes Wong as a veteran startup executive and Rachidi as a machine learning engineer.

Safeworld tests robots around people

The company’s work centres on simulated environments containing realistic human models. Its stated purpose is to evaluate robotic control systems, including the software that drives a robot. This approach allows scenarios involving people and robots to be examined in a simulation rather than relying only on physical encounters.

The source describes a factory blind corner as one setting where safety assessment may matter. A robot may need to approach a person carrying boxes, and its speed or stopping distance can affect whether a collision occurs. The simulation can represent a digital version of the setting, place a simulated robot under its real software, and run scenarios involving human models.

Investors fund an independent safety role

Shine Capital and a16z Speedrun led the seed round, which exceeded $12 million. Box Group, Carnegie Mellon University Endowment, Innovation Endeavors and SV Angel also invested. The source does not state the company’s valuation, the ownership terms, or how much each investor contributed.

The founders argue that robot makers may value a third party assessing their systems. The source notes that robot builders also use internal tools, while Safeworld’s founders point to their specific expertise and the potential value of external validation. No regulatory authority, legal status or formal safety standard is named in the reported announcement.

Simulations leave practical questions

The source gives tripping and falling as another scenario used in testing. Recreating such events with people in physical trials would require someone to fall repeatedly for the robot, whereas simulation offers a way to run those cases virtually. Zhao’s account also frames robot evaluation as difficult because robots operate in unstructured environments and safety requirements can differ between facilities.

The announcement does not specify which control systems will be assessed first, how the company will measure a safe outcome, or how customers will use its results. It also does not identify a regulator overseeing the work or describe an adopted legal requirement. Those omissions leave the precise relationship between Safeworld’s evaluations and any eventual industry standard open.

Safeworld’s stated next step is to develop evaluations using simulated robots and realistic human models. The company’s founders say an industry safety standard should be built while robots are being designed and deployed. Further details on the platform, customers and evaluation criteria were not provided in the source. The funding announcement therefore establishes the company’s aim and backers, while leaving its operating model and external validation process unspecified.