Comment by gshulegaard | Hacker News Reader

Comment by gshulegaard | Hacker News Reader

Parent Full thread

> What did the autopilot do and when did it do it seems to be missing from this round of reports on this incident and thats pretty lame.

I am pretty sure this is because it has been reported on pretty thoroughly.

https://www.tesla.com/blog/tragic-loss

https://www.nytimes.com/interactive/2016/07/01/business/insi...

The general sense is that the system is designed to apply brake when both radar and the camera system agree there was an obstacle (this may no longer be true since there was indication that decoupling was being worked on).

It has been suggested that the white trailer against a bright blue sky didn't register on visual detection. Thereby there was not a consensus between radar and visual that there was an obstacle so the brake was not applied.

> we would expect it to before broadly deploying these devices.

And this is why there is focus on the driver. What is reasonable expectation? Are there adequate warnings and documentation about the capabilities of the system? Autopilot is not marketed as an autonomous driving system...it (at the moment) explicitly says drivers should continue to monitor the system to manually intervene if required.

Further NHTSA research suggest as high as a 40% crash reduction since the wide introduction of Tesla's autopilot mode:

https://techcrunch.com/2017/01/19/nhtsas-full-final-investig...

This is a single accident...and arguably the driver was the final safety gate and they failed to apply corrective action (in addition to the automated safety systems). It does not appear as if the driver had sufficient reason to believe the system was capable of complete autonomous operation. And it does not appear as if there was a critical flaw or bug in the system that was putting consumers at risk.

And just briefly going over the items in the roughly 500 page report there does appear to be a lot of detailed examination of the systems.

Awesome reply, thanks.

I'm actually even more bothered by this.

This looks like a super common scenario people in cars face daily.

I acknowledge the technical obstacles to "wide view" forward sensors but wow, this system seems very very limited.

""“The limited time the target is in the field of view prior to impact challenges the system’s ability to perform threat assessment and apply the CIB system. A target is usually recognized very late or not at all prior to impact.”"""

It seems like the FOV of the sensors is not adequate.

The system should be looking backward to assess the danger of a sudden braking?

Sensor fusion should resolve ambiguity via probability... one line in the reporting suggested it needs both systems to "agree" which is a binary, this is probably just reporter confusion.

This system doesn't seem ready for anyone younger then 21 full stop.

If it doesn't radically improve it's going to have many more problems as it gets in the hands of a wider range of drivers beyond affluent people in middle age+.

These systems are "lane keepers" and really the marketing and driver education should be tightly focused around what they DO NOT do.

I genuinely look forward to less humans in control of cars but this is going to be a rough road.

There has been discussion about the limitations of the system...including a lot about how a Lidar system would have mitigated this. I think this underscores the fact that driver's must view the Tesla system as a collection of safety features rather than a fully autonomous system.

And I think you are correct that there is driver maturity required in order to exercise restraint/use the system responsibly given this and its potential abuse.

But a 40% decrease in crashes involving Teslas after the wide deployment of the Autopilot safety features would seem to indicate (to me) still that the overall safety of both Tesla drivers and the public around them is improved by its existence and availability.

> If it doesn't radically improve it's going to have many more problems as it gets in the hands of a wider range of drivers beyond affluent people in middle age+.

This is a very interesting and astute observation. While I question whether or not the current driver sample/population is materially more or less responsible than those that would have access to the upcoming more affordable Model 3, it is something worth monitoring (by the NTSB).

At the end of the day this is an evolving area. While this incident highlights the limits of Tesla's system, it is only one and there is evidence to suggest that overall safety is improved despite its limitations.

I believe Tesla rolled out an update shortly after the crash that fixed the 'bug' of it not recognising the white trailer against the blue sky.

> It does not appear as if the driver had sufficient reason to believe the system was capable of complete autonomous operation. And it does not appear as if there was a critical flaw or bug in the system that was putting consumers at risk.

I am going to disagree there. Tesla names its system autopilot. Uses the fact that autopilot that includes collision avoidance reduces accidents by 40%, and then claims that there system is safer than a human (assisted human safer than a human, genius!). Elon musk has a statement every month about self driving autonomous driverless cars. They even have the gall to say that new cars have "full self driving hardware", is proof of outrageous statements even required?

And all this hype/misleading advertising is fine because Tesla showed a few warnings? And the driver didn't have sufficient reason to believe the car was capable of complete autonomy. A simple Google search will tell you that this driver wasn't the only one.

Reply on news.ycombinator.com