Shivansh Inamdar: Pioneering Physical AI To Teach Robots Real-World Handling At Nimble Robotics

The ChatGPT Moment for Robotics

When will robotics have its ChatGPT moment? The question has become the technology industry’s favourite debate. Nvidia chief executive Jensen Huang declared at CES this year that “the ChatGPT moment for physical AI is here,” describing it as the point when machines begin to understand, reason and act in the real world. Prominent robotics founders counter that it remains years away. Trillions of dollars in projected market value hang on who is right.

The debate exists because acting in the physical world is a fundamentally harder problem than generating text. A chatbot that makes a mistake can apologise and try again. A robot arm reaching into a bin of unfamiliar products gets no such luxury. It must see the object, decide how to grip it, and move it correctly, at speed, thousands of times a day, whether it is a glass bottle or a plush toy, sitting upright or buried at the bottom of the bin.

“People keep asking when robots will get their ChatGPT moment,” says Shivansh Inamdar, an engineer who builds this technology in production. “Walk into an automated fulfilment centre and watch a robot pick a thousand different products it has never seen before, hour after hour, and you realise it is already quietly underway. The hard problem was never picking one object a thousand times. It is picking a thousand different objects, sight unseen, and getting each one right the first time.”

Inamdar is a senior robotics software engineer at Nimble Robotics, the San Francisco-based AI robotics company whose technology powers autonomous fulfilment for major American retailers and underpins FedEx Fulfillment, a partnership FedEx underscored by investing in the company. Trained at the University of Pennsylvania, where he earned dual bachelor’s and master’s degrees spanning physics and computer science, Inamdar joined Nimble in its early days and has spent over five years building the robotic arm manipulation software that lets its robots pick, pack, and sort merchandise with no human in the loop. Today those systems run across a fleet of hundreds of robots handling tens of thousands of items every day.

His work spans nearly the entire autonomy stack: perception systems that let robots interpret the hardest visual cases, grasp selection that combines learned models with physical checks, real-time motion, coordination software that orchestrates entire fleets, and a physics simulator that lets engineers validate changes without touching a robot. It is the unglamorous layer of physical AI: the difference between a demonstration video and infrastructure a retailer can bet its holiday season on.

Born and raised in India, Inamdar showed his instinct for building early. As a student, he co-founded Aarogya, a social enterprise that used software to widen access to medicines in India, and the venture won the University of Pennsylvania’s President’s Engagement Prize in 2020, one of Penn’s most prestigious honours, awarded to a handful of graduating seniors each year.

Why the Warehouse Is Where Physical AI Gets Real

The current wave of enthusiasm for embodied AI, from humanoid robots to foundation models trained on physical interaction, still lives mostly in demos and research videos. Inamdar operates a step beyond that conversation: taking the same frontier techniques and putting them to work in production, solving real-world problems at commercial scale today.

Warehouses, in his view, are where physical AI is being forced to grow up. The environment is semi-structured, the economics are unforgiving, the opportunity is enormous, and the diversity of objects, spanning millions of products in every shape, material, and packaging, makes it the perfect proving ground for general-purpose manipulation. A system that handles anything a retailer sells is a system learning the skill every future robot will need, whether it works on a factory line, in a hospital, or in a home.

“Reliably handling an object you have never seen is the central unsolved problem of physical robotics,” Inamdar says. “Warehouses are simply where that problem is being solved commercially first. The capability being built there is the capability the rest of robotics will stand on.”

That gap between laboratory success and production reliability is where he has focused his career, and where, he argues, the field’s real progress is measured. Academic systems that succeed most of the time are impressive; commercial systems must approach the reliability of human workers, every hour of every shift.

A Voice in the Field

Inamdar is an active voice in the research community working on exactly this transition. In the past year alone, his technical work on robotic perception and grasping has been presented in Vienna, Denver, and Sydney, at workshops convened by the world’s leading computer vision and robotics conferences. He holds advanced memberships in engineering and leadership associations, and serves as a judge for international technology and AI award programmes, evaluating the next generation of AI products and startups. He also advises early-stage founders in both India and the United States.

The Long View

Ask Inamdar where physical AI goes next, and he points not at humanoid demos but at quiet ubiquity: manipulation reliable enough that it disappears into infrastructure, the way logistics software already has.

“Software finally has hands,” he says. “The interesting work now is teaching it to use them carefully enough that people can depend on it, for their packages today and for much more than that tomorrow.”

The engineers who can do that work, with an understanding of both the physics of the world and the software that must reason about it, remain rare. The ones shaping the field are those who understood early that the future of AI would not just be written. It would be built, gripped, and moved, one object at a time.