Comment by gvkv | Hacker News Reader

Comment by gvkv | Hacker News Reader

I think a more fundamental problem here is that self-driving cars are as much an infrastructure problem as they are a technological one.

As an analogy, consider hybrid vs electric vehicles. In places like North America with large, open spaces, electric vehicles really only serve a specific type of urban driver. The culture, infrastructure and geography dictate 600km distances which really aren't practical at the moment with current battery tech. Whereas hybrid vehicles can (or could) quite easily reach that range with options to recharge once you get to your destination or have a longer stopover and still use existing infrastructure. The focus on purely electric is a lost opportunity for anyone who needs power or long distance.

Similarly, cars could be designed to be self-driving in the easy cases; highways, certain urban thoroughfares, particular times of day and the like where existing vehicle and pedestrian flow patterns eliminate the edge cases. coordinating systems along the aforementioned types of roads could be installed as was done for cellular service and GPS and other protocols could be developed to ensure safety and reliability as well as fallback in case of emergencies.

Instead, we've decided on all-or-nothing bets which don't move things forward--or at all--and my worry now is that we'll lose an opportunity to pick the low-hanging fruit and solve the harder problems incrementally over time.

But chasing the higher hanging fruits might allow for breakthroughs that you would not see if you only went for the low hanging fruits.

The range anxiety is less and less problematic with EV. The Tesla Roadster 2 already is said to have a range of more than 1.000 km. Add current research in the fields of solid state batteries and super capacitators and you have the possibility to reach those numbers even with less expensive versions of EV. German automakers already calculate that by 2026 electric engines will be cheaper and more capable than their ICE counterpart.

If you go for that easy middle ground like hybrid cars that you suggest, you limit yourself to the local maximum of that solution. Hybrid cars have the same maintenance cost as non hybrid cars and additionally the complexity of balancing both engines. The only saving in maintenance cost is by going full electric. In the same way you might only achieve certain breakthroughs by actually going for full autonomy even if it wont work perfectly for the next decades for all edge cases.

The choice is about releasing version 1.0 vs constantly adding features that are better suited for versions 2, 3, 4+. What can be done now, at reasonable cost and technology level? My position is that we could have a system right now that would alleviate traffic congestion and offer greater safety. Advancement on the harder problems comes with real-world experience in the field.

The doing is the learning.

Bloomberg believes Teslas numbers are a little optimistic and claim it's not possible with "current" battery technology: https://www.bnnbloomberg.ca/tesla-s-newest-promises-break-th...

It doesn't mean Tesla won't do it. But it will be a big deal if they do.

Agreed. We can more easily follow the autopilot patterns of commercial flights such that cars can activate autopilot once they have officially entered freeways, and disengages again after exiting.

Doing so can also help spur further investment into the Interstate system, which as an immigrant was one of the best inventions that America had made in creating a higher quality of life than other countries.

Hybrid drivetrains (tiny engines with forced induction and electric motors as a blueprint) plus driver assistance tech is a much brighter near future than trying to wrangle busy city streets and electric charging.

Could you do it without all or nothing? Mixing self-driving cars with humans means the intelligence has to understand irrational behaviour and has to respond appropriately (speed up rather than slow down, ignore the potential threat, account for traffic behind you), which is sometimes counter-intuitive. And it comes in many ways.

Will a self-driving car know that you don’t pump the brakes when sliding on snow? What about hydroplaning?

If that didn’t exist we wouldn’t have car fatalities or accidents.

I don't think any of those examples are edge cases. The first set are normal traffic conditions that in the context of self-driving cars are easy to solve, especially in narrowed conditions such as on a highway. Moreover, mid-range cars already have collision warning and automatic braking systems. As to your example of traction issues, pretty much every modern car that I'm aware of has had computer assisted traction control systems for a while now.

The edge cases that are difficult essentially boil down to entity recognition; unexpected and moving obstacles, road sign changes, traffic light outages or alternate signal pathways and the like. Some of those definitely would require government level coordination which is about a lot more than technology.

Reply on news.ycombinator.com