In this episode of the Sunny Ray Show, host Sunny Ray speaks with Shivoh Chirayil Nandakumar of RideScan AI about physical AI and what it takes to make robots trustworthy beyond the lab. Sunny discloses that he is CEO of the public company Humanoid Global, which holds a minority stake in RideScan. Shivoh describes RideScan as a Fitbit for robots, an independent monitoring layer that scores robot reliability. He traces his motivation to his PhD work in AI robotics in Edinburgh, where a robot confidently misidentified a table as a box, and to his paper on anomaly detection and aleatoric uncertainty. The conversation covers real-world failures, why robot self-reports are not evidence, why monitoring must be independent of manufacturers, and what pacing means as robots approach a ChatGPT moment. Shivoh also explains how RideScan calibrates to a task, generates a risk score per execution, works without cameras, and how a first customer paid upfront after a missed robot error.
A Fitbit for robots: why physical AI needs independent monitoring before mistakes in the lab become harm in the real world.
What is RideScan building, and what problem are you solving?
I usually call RideScan a Fitbit for robots. The idea is to independently monitor autonomous robots so we can make sure they are safe and reliable when they are deployed in public spaces or in any industry.
What first pulled you toward robots, and why did trust become your problem to solve?
During my PhD in AI robotics in Edinburgh, a robot I was training sometimes picked up a table and confidently reported it had picked a box. Mistakes in the lab are fine, but not in the real world. I published a paper on anomaly detection and aleatoric uncertainty, which led me to build detection.
When did you realize a robot's own report about itself isn't evidence?
It was while programming AI integrated robots in the lab. A robot completes its defined task and doesn't care about minor mistakes, like dragging a cable attached to sensitive equipment. Those small errors can have high consequences, and only external observation can catch them.
What does pacing mean for physical AI?
Frontier AI labs talk about pacing so growth stays controllable and beneficial. Robots are nearing a ChatGPT moment, handling flexible tasks like dishes and laundry. That is useful, but more flexibility means more potential for mistakes, so we should solve those problems proactively.
Why does the check have to be independent of whoever built the robot?
If manufacturers monitor themselves, each uses its own metrics and can claim its robot is better. It is like writing my own exam, passing it, and telling an employer I'm good. Without independent checks, trust cannot be foundational. Other industries have independent structures, and robotics will need them too.
Where do predictive maintenance and real safety assurance begin?
Predictive maintenance serves robots already in production, where failures at random times cause downtime. Safety becomes more important with flexible systems like humanoids, drones, mobile platforms, and robotic arms working near humans, where predictable behavior matters and any variation should be flagged early.
What happens when a customer starts with RideScan?
RideScan is for enterprises commissioning robots. It learns what the robot is supposed to do, then generates a risk score for every execution. That reliability score supports predictive maintenance, safety, and better decisions. We recently deployed on a German robot in under two hours.
How did your first customers find you, and why did they say yes?
I showed a demo to the CTO of an integrator that sells robots to enterprises. Two months later he called: a deployed robot had made a mistake their software didn't show. They paid upfront because they saw RideScan could independently capture that behavior deviation.
Building something daring? Sunny talks to founders like this every day. Fifteen minutes to see if your story belongs on the stage.
Claim your pre-interviewSend a founder here and we will know it was you.