In an air-conditioned office the room is cool, but in real-world scenarios you don't control the temperature, so overheating becomes more severe. Laundromats don't necessarily have the best Wi-Fi, so running models over the cloud and uploading data both slow down. And there's the operation challenge of trusting a robot on a customer site without something going catastrophically wrong.
More from this episode
Jason explains what they didn't realize: in the office they had air conditioning so the room was cool, but in a lot of real-world scenarios "you do not have control of the temperature, so overheating becomes more severe." You also don't have good control of the network — "laundromats don't necessarily have the best Wi-Fi," so if you're running models over the cloud, Wi-Fi becomes a bottleneck, and uploading data from on-site collection becomes a lot slower. Then there's the operation challenge: "how can we trust to put a robot in your real customer site and not worry about something goes catastrophically wrong" — for instance a robot folding napkins in the back office that you don't want to catch on fire by accident. His takeaway is that the hard part of robotics isn't just getting AI to work, you also have to make sure the deployment flywheel goes smoothly.