Accepting liability doesn't mean they require the system to be perfect. Accepting liability really just comes down to who's insurance company pays when it fail

Accepting liability doesn't mean they require the system to be perfect. Accepting liability really just comes down to who's insurance company pays when it fails. If the manufacturer can demonstrate to the insurance company that their self driving system fails less frequently (or lower overall cost of claims payouts) than human drivers then the insurance company will happily charge a lower premium for cover. Even if it's not perfect. The closer the system can be demonstrated to be to perfect the lower the insurance premium is going to fall. There is certainly a regulatory moat being set up here as it will prevent new startups from entering the space unless they can find an insurance company willing to cover their risk profile, which in turn means demonstrating a suitably high level of safety to keep the premiums affordable. But I think in this matter I want a level of regulation that says if you're making something that's used on public roads, and risks injuring or killing people you'll accept liability when it goes wrong. It's not much different from the system of mandatory insurance for drivers in most countries. If you're going to drive on the roads, you must meet a minimum insurance level whereby you can accept liability for any failure that causes damage to other property or people.