[kidmenot]

> I can't imagine partnering with a traditional automaker will increase the pace.

Right? I was thinking exactly the same. One would think that having a car built in-house enabled them to iterate that much quicker. Then again, Google is full of smart people, so there surely is a reason behind this decision.

[michaelt]

  One would think that having a car built in-house
  enabled them to iterate that much quicker.

There tends to be a pretty clear separation between the self-driving sensors and brains ('the new stuff') and the car platform ('the old stuff').

As long as you're willing to have some sensors and actuators visible (i.e. you're working on a prototype rather than a consumer-ready product) there's not much technical value added from building a custom car vs modifying an existing one. Indeed, it can easily slow you down as now you have a bunch of work to do developing a new car.

The car that won the DARPA grand challenge was fairly close how it came off the production line - except for an off-the-shelf electrical actuator system developed for disabled drivers, an off-the-shelf secondary alternator to power all the sensors and computers, and a bunch of sensors bolted to the roof rack.

The reasons to build your own are business/PR rather than technical/development speed. For example they can make it look cute and unaggressive which might be good PR; or insist on a public transport style vandal proof design for an on-demand business model. These might be good reasons, but they're business reasons not technical reasons.