Perseverance has driven itself across Mars for three years now, covering roughly 27 kilometers with 90 percent of that distance completed autonomously.
The rover receives commands from Earth with a 20-minute communication delay each way—by the time humans see what the rover saw, decisions already made themselves. This sounds like progress, like artificial intelligence finally doing what we've promised.
But it is actually the return of a system we abandoned fifty years ago because it didn't work. In 1971, astronauts drove the Lunar Roving Vehicle on the Moon, steering it themselves across 35 kilometers during Apollo 15 and Apollo 17, making real-time decisions based on what they saw and felt.
What killed the rover wasn't the concept—it was the crashes. The LRV flipped in soft regolith, got bogged down. Required a trained human pilot constantly correcting in real time to work at all, and even then it barely worked. So we stopped doing that. We built rovers that planned their own routes, that checked their own systems, that made thousands of micro-decisions without waiting for home.
The LRV needed a pilot. Perseverance needs an engineer who understands its constraints well enough to tell it where to go and trust it to figure out whether it can get there safely.
Sojourner in 1997 was slow and constrained, but it worked alone. Spirit and Opportunity expanded what autonomous rovers could do. By the time Perseverance rolled out, autonomy wasn't a feature—it was the only option. The real innovation isn't that Perseverance drives itself. It's that we've built enough computational redundancy and sensor precision into the rover that it can make good decisions without a human in the loop at all.
That distinction matters when you're designing systems for Earth. Self-driving cars promise the same thing Perseverance delivers. They're still waiting for the moment when we trust them enough to let them fail alone. We haven't actually reached that trust yet.