Why we're building SafeWorld—and who we're looking for

A robot stops when a person crosses its path. Move a cart into the scene, and the test changes. Part of the person disappears from view. The robot may need to respond before the person fully emerges.

That small change raises several questions. Did the sensor capture enough information? Did the software detect the person? What did the robot do next? Can another engineer repeat the test and reach the same conclusion?

These are the kinds of questions behind SafeWorld. We're building simulation and testing tools for robots that work around people. We want engineering teams to investigate difficult situations and understand the limits of their evidence.

The people behind the work

SafeWorld brings together research, software engineering, and company-building experience. Our co-founders are Ding Zhao, Kyle Wong, and Simo Rachidi.

Ding Zhao

Kyle Wong

Simo Rachidi

Ding Zhao, Chief Scientist. Ding has spent nearly two decades working on the safety of autonomous systems. He leads the Safe AI Lab at Carnegie Mellon, and is an NSF CAREER Award recipient in this space. Few people have thought as deeply about how to make increasingly intelligent machines operate safely in the physical world.

Kyle Wong, Chief Executive Officer, brings experience building companies, including founding Pixlee. His focus includes understanding customer needs and building a company that can serve them.

Simo Rachidi, Chief Technology Officer, specializes in machine learning, anomaly detection, and large-scale data systems. At Salesforce Einstein, his work included pipelines and models in systems processing over 10 billion events per day. He helped build an early internal retrieval-augmented generation system for cybersecurity and modernize the supporting stack. Earlier, he was a founding engineer at Pixlee and a co-founder of Distinc.tt. His master’s degree focused on autonomous driving.

Those perspectives meet in a practical question: what does a customer need to decide after running a test? The answer should shape the software we build and the evidence it produces.

The engineering problems we care about

Human movement is one part of the work. A person may stop, turn, bend, or walk behind something that blocks a sensor. Representing those situations requires choices about motion, geometry, timing, and behavior.

Each choice has consequences. A fixed trajectory can help reproduce an encounter. A person who reacts to the robot requires a different model and a different evaluation. We need to make that distinction clear to the engineer using the result.

Another challenge is keeping comparisons meaningful. If lighting, motion, sensor settings, and software all change, a better score can be difficult to explain. Tools should help users control those changes and preserve the context of a result.

Then there is the work around the simulation itself. Inputs need clear formats. Failed runs need useful explanations. Results need to remain connected to their configuration. An engineer returning to a test weeks later should be able to understand what happened.

These problems cross several disciplines. They need people who care about robotics and people who care about reliable software. They also need people willing to question whether a technically interesting feature helps the customer.

How we want to work

Our engineering principles start with ownership. Find the problem, make its scope clear, and carry it through to a useful result. Ask for help when it improves the outcome.

We value curiosity with a practical purpose. When a result looks wrong, investigate the input, assumptions, and measurement before explaining it away. An unexpected result can teach us where a tool needs to improve.

We also value small, inspectable changes. A focused test or a clear failure report can move a project forward more than a broad claim. Reliability, documentation, and evidence are part of the work.

Disagreement should be direct and respectful. Explain the concern, show the evidence, and stay open to a better answer. That matters when the subject is safety and the available information is incomplete.

Who will enjoy this work

You may enjoy SafeWorld if you like moving between a technical problem and the person who needs it solved. You might care about human motion, perception, simulation, machine learning, or the software supporting those systems.

We are especially interested in how you reason through uncertain results. What did you assume? How did you test it? What changed your mind?

A project with a clear explanation can tell us a great deal. Show what you built, the choices you made, and the limits you found. The work ahead includes plenty of problems whose answers are still open.

Open roles

Explore the roles below to learn about each team's work and application process. Choose the role that best matches the problems you want to solve.