The Pulse
SafeWorld Raises $12.2M to Test Robot Safety in Simulation
SafeWorld has raised $12.2 million in seed funding to simulate how AI-controlled robots behave around people in dangerous, unexpected situations. The Palo Alto startup says it is already running early pilots with automotive, warehouse autom
AI.info Team · Written with AI ·
Robot makers often manage uncertainty with physical barriers and speed limits; SafeWorld argues that simulated tests can expose more hazards before machines reach people. The Palo Alto startup announced October 5 that it has raised $12.2 million in seed funding to build safety-testing software for robots, including systems controlled by generative AI. Its pitch is ambitious: model risky encounters in virtual environments rather than rely on physical trials alone.
But simulation does not settle the question of whether a robot is safe in a real workplace. SafeWorld is still deciding whether to sell its technology as a self-service platform or provide testing as a service, and the company’s case for adoption rests partly on early pilots and its promise to run tests repeatedly as robots change.
From blind corners to simulated falls
SafeWorld says customers can create scenarios in a browser using past incidents, safety standards and robot logs. The software then tests the robot across thousands of variations, with simulated people moving in realistic ways. The company says teams can repeat those tests after software updates or when a robot enters a new environment.
In an interview with TechCrunch, co-founder Kyle Wong described blind corners, stopping distances and workers carrying boxes as scenarios that matter in factories. A simulation could also test how a robot responds to a person who trips or falls—an event that would be difficult and unsafe to reproduce repeatedly with a real worker.
SafeWorld says its technology uses generative 3D simulation and reactive human-motion models. Its launch announcement says companies can run millions of edge-case simulations, while its description of the test process says a robot can be run through thousands of variations. Those are company claims, not evidence that simulation alone can establish safety in every physical setting.
Investors back an independent testing case
Shine Capital and a16z Speedrun co-led the seed round. Box Group, Carnegie Mellon University Endowment, Innovation Endeavors and SV Angel also invested, alongside additional funds and individual backers named in the company announcement. SafeWorld says it is already running early pilots with automotive manufacturers, warehouse automation companies and medical-device makers.
The company was co-founded by Carnegie Mellon University Safe AI Lab director Ding Zhao, Kyle Wong and Simo Rachidi. Zhao argues that testing must account for the people and conditions robots meet after demonstrations end. “It is not the robot in the vacuum, in the demo, that we are worried about. It is the robot that is deployed at scale, with people who potentially never operated a robot before,” he told TechCrunch.
Simulation still has to meet the worksite
One of SafeWorld’s named partners, Gritt Robotics, is developing robots that help workers install photovoltaic panels at industrial-scale solar farms. Its CTO, Vishal Dugar, told TechCrunch that proving a system safe through formal mathematical methods is difficult and that safety “necessarily has to be done empirically.” The partnership gives SafeWorld a real deployment context, but it does not by itself show how well simulated results predict robot behavior on a working site.
That gap matters because robots operate in varied places with different procedures, obstacles and human behavior. SafeWorld’s independent-testing argument is that robot makers may benefit from outside evaluation and shared safety cases, rather than relying only on internal testing. Yet TechCrunch reported that the startup had not settled whether its main offering would be software customers use themselves or a service run by SafeWorld’s team.
The $12.2 million round gives the company room to develop its testing tools and pursue those pilots. The immediate measure of progress will be whether simulated scenarios help customers make safer deployment decisions—and whether those results hold up when robots work beside people.