Navigation is the hardest and most important thing

It’s tricky to build and program a robot that can navigate through your home without getting stuck or lost. Bruno Hexsel, a former software engineer for Neato, told us that he spent a huge chunk of his time at the company working on algorithms to help robots get unstuck from common hazards.

It’s not so different from a self-driving car. The stakes are much lower with a 9-pound robot blooping around your living room than they are with a 3,500-pound car hurtling down a highway. But both need to navigate their environments on the fly, with an unpredictable set of obstacles and hazards that can change constantly.

The challenges are similar enough that several former robot-vacuum engineers—including Hexsel—have subsequently worked in the autonomous-vehicle industry. Duane Gilbert, formerly of iRobot, has worked for a self-driving car company called Nio.

Gilbert told us that it’s arguably harder to make a robot vacuum navigate than a car, because your home has more variables than a road. Floors don’t have traffic lines or mile markers or light poles to orient the bot. Different flooring surfaces have different traction, whereas asphalt is pretty consistent. And you can put more sensors, of a higher quality, in a $40,000 car than you can in a $500 robot.

So what does a robot vacuum need to do in order to navigate a real home?

First, don’t get stuck. Robots that keep moving are the ones that keep your floors the tidiest, so it’s vital that they don’t get hung up, tangled, or otherwise obstructed in the middle of a job. If you set your bot to clean while you’re out of the house, but it gets stuck under your dining room table within the first 10 minutes, you’ll still have dirty floors when you get home.

The most common bot-trapping hazards include power cords, charging cables, stray laundry (especially socks), curtains, bed sheets, floor-to-rug transitions, rug fringe and tassels, floor registers, tall thresholds (some bots can’t get over them), and furniture with tube-shaped or extra-wide supports that lay across the floor. Higher-end models can have trouble in dim lighting (if they have camera-based navigation) and with chrome furniture legs or mirrors (if they have LiDAR-based navigation). Black rugs (or other dark, non-reflective flooring) can look like a bottomless pit to a bot’s anti-drop staircase sensors, and many models won’t clean them. Some homes have more of these traps than others; most homes have at least a few. Great bots work well around most of those hazards, and some struggle with several of them. We’ve even seen some poorly programmed bots just give up and quit when they have to deal with too many obstacles at once.