In this episode of the Ready Set Do podcast w/ Naman Pandey, featured not-expert is Anshuman Kumar. Anshuman is the head of hardware at Matic Robots - whose $1100 smart home robot has been rated by Wired magazine as the best robot vacuum they've ever tested.

All my links: readysetdopodcast.com

What does it really take to build the best robot vacuum in the world—the one that Wired Magazine called “the most impressive cleaner we’ve ever tested”? In this episode I sit down with Anshuman Kumar, Head of Hardware at Matic Robots, to pull back the curtain on seven years of engineering decisions, failures, and counter-intuitive bets that turned a side-project into a category-redefining $1,100 smart-home robot.

Around the ten-minute mark you’ll hear Anshuman roast the entire robot-vacuum industry: he argues current machines “suck” not because suction is weak, but because their perception is worse than a toddler’s. That discussion tees up Matic’s first “aha” moment: teaching the robot to understand rooms the way people do. Leveraging semantic segmentation and a hefty NVIDIA Orin on board, Matic can remember that the “kitchen” is a grease-prone area, or that the dog’s food bowl is an immovable obstacle, and then generate the right cleaning plan without constant app babysitting.

Of course, a smarter robot raises thorny privacy questions. Mid-episode Anshuman outlines the team’s decision to keep all vision processing on the robot, never in the cloud—no images leave your home, period. That choice forced some brutal trade-offs in PCB layout, battery mass, and thermal engineering, but it paid dividends in user trust and latency-free autonomy. He also details how on-device compute enables features like “clean my bedroom before 7 am but skip the nursery,” entirely offline.

Then we get into the juicy stuff: the four most counter-intuitive hardware decisions Matic made and why each one broke conventional wisdom but unlocked breakthrough performance:

1. Sacrificing sheer suction for airflow geometry—because debris intake angle matters more than raw Pascals.

2. Discarding low-profile design; the robot stands noticeably taller to house a larger optical assembly and lift its camera above clutter, boosting coverage accuracy by 30 %.

3. Choosing a louder motor profile at specific frequencies that human ears quickly ignore, trading “whisper-quiet but high-pitch” for a hum that blends into background HVAC noise.