Google says it has turned a corner in its pursuit of a car that can drive itself.
The leader of the technology giant's driverless car project wrote in a blog post that test vehicles are becoming far more adept at city driving.
They can already comfortably handle motorways, he said, but city driving presents a virtual obstacle course of pedestrians, cyclists and blind corners.
Google says the cars can now negotiate thousands of urban situations that would have stumped them a year or two ago.
To navigate and avoid crashes, Google's fleet of retrofitted Lexus SUVs relies on sensors such as lasers and radar. A driver is ready to take over if needed.
Google has said it wants to get the technology to the public by 2017.
"We're growing more optimistic that we're heading toward an achievable goal - a vehicle that operates fully without human intervention," project director Chris Urmson wrote.
His post was the company's first official update since 2012 on progress towards a driverless car, a project within the company's secretive Google X lab.
In initial versions, human drivers would be expected to take control if the computer fails. The promise is that eventually there would be no need for a driver. Passengers could read, daydream, even sleep or work while the car drives.
Google maintains that computers will one day drive far more safely than humans, and part of the company's pitch is that robot cars can substantially reduce traffic fatalities.
The basics already are in place. The task for Google and traditional car firms, which are also testing driverless cars, is perfecting the technology.
Sensors create 3D maps of a self-driving car's surroundings in real time, while Google's software sorts objects into four categories: moving vehicles, pedestrians, cyclists and static things such as signs, kurbs and parked cars.
Initially, those plots were fairly crude. A gaggle of pedestrians on a street corner registered as a single person, for example.
Now, the technology can distinguish individuals, according to Google spokeswoman Courtney Hohne, as well as solve other riddles such as construction zones and the likely movements of people riding bicycles.
To deal with cyclists, engineers initially programmed the software to look for hand gestures that indicate an upcoming turn. Then they realised that most cyclists do not use standard gestures and others weave down a road the wrong way.
So engineers have taught the software based on thousands of encounters during the approximately 10,000 miles the cars have driven autonomously on city streets, Ms Hohne said.
The software projects a cyclist's likely movements and plots the car's path accordingly, then reacts if something unexpected happens.
"A mile of city driving is much more complex than a mile of freeway driving, with hundreds of different objects moving according to different rules of the road in a small area," Mr Urmson wrote.
Before recent breakthroughs, Google had contemplated mapping all the world's stop signs. Now the technology can read stop signs, including those held in the hands of school crossing wardens, Ms Hohne said.
While the car knows to stop, just when to start again is still a challenge, partly because the cars are programmed to drive defensively.
At a four-way stop, Google's cars have been known to wait in place as people driving in other directions edge out into the intersection - or roll through.
The cars still need work on other predictably common tasks. Among them, understanding the gestures that drivers give one another to signal it's OK to merge or change lanes, turning right on red and driving in rain or fog, which requires more sophisticated sensors.
"You can count on one hand the number of years until people, ordinary people, can experience this," company co-founder Sergey Brin said in September 2012. He spoke at a ceremony where legislation to make the cars legal on public roads in California was signed.
To date, Google's cars have gone about 700,000 miles in self-driving mode, the vast majority on motorways, the company said.
Google has not said how it plans to market the technology. Options include collaborating with major car firms or giving away the software, as the company did with its Android operating system. While Google has the balance sheet to invest in making cars, that likelihood is remote.