Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Flying in uncontrolled airspace in VMC is a “see and avoid” environment, meaning this looks like a pretty bad screw up by Amazon.

The fact that two different drones crashed into the same object raises even more serious questions on the quality of Amazon’s tech and their ability to safely monitor it.



Two drones doesn't really mean anything if they were following a similar flight plan to make a delivery at the same location right?


It means it wasn't a fluke or a bug specific to one drone, but something wrong in the overall software approach.


The repetition strongly indicates it’s a bug. No reason to think it points to a fundamental flaw in the approach.


i mean they both flew into invisible drone traps.

I don't know that amazon engineers should be expected to see e.g. a moving small steel cable under tension.

That and customers are required to select safe delivery drop zones.

I would like to see better "oh shit we're crashing let's try not to kill anyone" protection, e.g. research on improving controlled landings on damaged drones. Maybe refusal to deliver if there are any detected humans in the drop zone (which may well already exist).


It means Amazon’s approach to its “see and avoid” responsibility is fundamentally flawed in some way vs this being a one-off fluke with a broken sensor or other anomaly.


ok but also two drones crashing isn't "more" of a problem than one drone crashing really.

Sounds like the anamoly here was a very unsafe landing zone (which is outside the customer agreement as it happens).

small steel cables take out human pilots too..

Would be really curious how they might guard against adversarial drone deliveries. Kinda weird to have end users basically piloting your $100K (I'm guessing) vehicles.


At the same time? If there's a crash there should be an automatic system which geofences off that area making it impossible for other drones to go near there, while the situation is assessed.


If a drone crashes, obviously no other drones should fly there until a human determines what went wrong and presses the 'resume' button. The fact that that system did not exist is a systemic problem.


That sounds like an engineer-week of work not really a ground-up systemic problem. But fair criticism.


The systemic problem is that they didn't spend the engineer-week on it. It's only an engineer week. That pays for itself after avoiding a single drone crash to say nothing of avoiding a second lawsuit.


well at least it's consistent


second time was because dev rejected bug report without QA replicating it


this is actually hilarious because now they can't call it a fluke or an act of god


Crashing it a second time is, in my books, an act of God. I guess God isn't really that fond of Amazon.


News Media: "Bezos has more money than God"

God: "Hold my staff"


> see and avoid

Wires are somewhere between hard and not possible to see, visually. The "fix" for this might be "that kinda looks like it might be construction over there, go around".


Even very small and cheap lidar can see cables just fine.

I work with 3d scanning lidar every day and I know this as a fact.

They have no excuse there.


In the sun, at 50mph? 50mph sets the minimum range of the scan (related to the "full stop" distance and fully loaded maneuverability), which sets the acquisition rate and angular resolution requirement to see a thin cable at that distance. I suppose maneuverability would also set the FOV requirement.

Are there any commercial drones that do it "right", with LIDAR?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: