No, self-driving cars aren’t here yet. But they are roaming a few select sections of American road. Waymo just launched a limited service in metro Phoenix (albeit with a safety driver behind the wheel); General Motors’ Cruise is testingin San Francisco; Ford is noodling around Florida; Auroraand Argo (which is closely aligned with Ford) swing through the hills of Pittsburgh.

And in many of these places, you’ll hear the same complaint from the humans sharing those roads: Man, these robots aren’t great at merging. I heard it during a recent trip to Phoenix, and reporters have found other (human) drivers whinging about too-timid AVs.

They’re not wrong: Self-driving cars do find merges challenging. In fact, it’s a problem that points to one of the harshest realities facing everyone teaching computers to drive themselves: For the foreseeable future, they’ll be sharing the road with human drivers. Humans who tend to favor flexible, communication-based interaction over following rules by the letter. Humans who can be capricious and inattentive. Humans who can be jerks, humans who can be generous.

“Merging is one of these beautiful problems where you have to understand what someone else is going to do,” says Ed Olson, the CEO and cofounder of the autonomous vehicle company May Mobility. “You’re going to get in the way of someone’s plan, because everyone wants to keep going straight and not let you in. You need to change their mind and change their behavior, and you do that by driving in

towards where you want to merge and they finally acquiesce.”

To understand the problem, think of the last time you joined the flow of traffic on the highway or a busy road. Maybe, just as you pulled up, you found a nice big gap to slide into. But maybe you weren’t so lucky, and you had to gauge the speed of the other cars in the vicinity to find a spot. You noticed the minivan speeding up to get ahead of you, and the sports car letting off the gas to make some room. You adjusted your own velocity and nudged to the left, communicating your intentions.

This is the kind of person-to-person ballet that’s hard to get a robot to perceive and understand. Getting it to hop on stage and join the company is even harder. That’s partly because merging calls on every bit of a self-driving system—and has a knack for revealing what’s not working.

“When a self-driving car is executing something like the merge, it has to identify where it is, what the the intentions of the drivers around it are, and then perceive and track them through time and space for the opportunity to slot in,” says Sterling Anderson, the cofounder of the autonomous vehicle startup Aurora. If those calculations are off by touch, the robot can muck up the whole maneuver.

(Anderson adds that piloting through an intersection is much more challenging, self-driving-wise, because the vehicles need to localize, identify, and then predict how a whole host of street actors will move. Not just other cars but less predictable players, like cyclists and pedestrians. At least on the highway, you’re only worried about other four-wheeled things.)

Which is another reason today’s self-driving cars aren’t great mergers, at least from a human perspective: There’s little to gain from being too aggressive, and much to lose. Smashing into a car on the highway? Bad. Annoying the the drivers behind because you’re being too timid on the road? Much less bad.

“Think of a self-driving car like someone with a student driver sticker on the back of their car,” Anderson says. You understand and tolerate their less-than-optimal skills because, well, they’re learning. So be patient with the novices in your midst, be them flesh or machine. And maybe consider making some room on the road.