I love this sort of thing in theory, but I would never trust my life to an electronic helicopter pilot.
I believe the tech can be made to be safe and reliable during normal flight. However, all the edge cases here are potentially fatal.
How well can we train a computer to choose a safe emergency landing spot and auto-rotate down if the engine fails? How will it cope with avionics/electrical failures? Or a geomagnetic storm?
How well will the system be hardened against malicious attacks? How well can we endow it with the judgment to avoid icing at low temperatures?
The answer to most of these questions might well be “pretty darn well, actually;” however, emergencies happen, and in those situations, I— and, I suspect, many others— will only ever feel safe with an experienced, context-aware, human pilot at the controls.
Well, probably by training the net under such conditions. Actually I see no reason that, with a sufficiently large and accurate dataset, a neural net can't be trained to pick out a "safe[enough]" emergency landing spot. And, a well trained net should be able to fly in any condition, including limited avionics, better than a human.
You might ask where this data will be coming from. I suspect 3D physics simulations are accurate enough that they could be used to generate data. Particularly if optimized for the physics behind helicopter handling.
Right now if you take an ordinary ambulance and let it enter one of Elon Musk's Boring Company tunnels and speed towards the hospital at 150 mph with no time outs for traffic that would be a large enough advance for most people with way less risk. Sometimes one new "good enough" technology overtakes a sexier one.
Would a sufficiently good track record convince you? Say, a million flights with 80% lower accident rates than human pilots?
A strict utilitarian should only care about accident rates. But as Greene [0] and others show, most people don't make moral judgements in a strictly utilitarian way. Thus, we judge homicide as worse than choosing not to save someone from death, even though in both cases someone dies as a result of a decision.
One of the lessons from probing moral calculus is that nobody can really articulate reasons for their actual moral decisions. People can articulate utilitarianism, but not what people actually do.
But, if you can articulate why you might prefer a human pilot over an autopilot, even given sufficient statistics that the autopilot is safer, it'd be interesting to hear.
The current helicopter accident rate for emergency medical service is 3.9 / 100,000 flight hours [0], so a million flights is probably large enough. But feel free to answer the question based on some larger number that would convince you it's statistically safer.
I think I could be convinced, eventually— but it would take quite a lot of evidence. A consistently lower accident rate than human pilots would be a good start, but it would be even better if those flights were conducted in the most punishing weather conditions possible and in all sorts of dangerous scenarios where pilot error is likely.
One of the biggest factors is that it is terrifying to imagine falling out of the sky in an aircraft under the command of a software-based pilot with which I cannot communicate and that I cannot see doing everything it possibly can to avert disaster. It may not be a rational fear, but we humans are not rational creatures.
It’s an issue of trust, which the folks interviewed in the article do acknowledge, to their credit. We can instinctively judge whether or not to trust another human being, because we have a lifetime of practice interacting with them. The burden of proof of safety for an uncommunicative mechanical system is much higher, and lots of testing (that I can observe on video) will go some way towards to building that trust.
Furthermore, things like avoiding thunderstorms, the motion of which we still cannot predict with computers (turbulence being an unsolved problem in physics), could be very tricky for an automated system to handle.
As autonomous technology progresses and we accumulate data, I might begin to feel differently. But for right now, without knowing all of the details of the system, I have no way of knowing how it is reasoning and perceiving the situation (I’m very familiar with the human consciousness, as I possess one myself) and this leads to a fundamental sort of uneasiness in my lizard brain.
I’m still bullish on technology, though, and this is an area in which I would be enthusiastic to be proven wrong.
an autonomous helicopter startup doesn't have unions with collective bargaining agreements (contractually obligating them to have at least 1 human operator).
And when then insist you take a helo ride for that cut on your wrist, you will be amazed at the $5,000 bill added onto your hospital bill and hope to god its covered by your plan. And to pay for this everyone will have to use it, say good by to the less expensive ground based EMS. US health care is great at replacing the less expensive stuff for the only slightly better, but vastly more expensive.
Contracts aside there’s a real issue: the human is there primarily to deal with other humans and all the random issues that pop up (which, as a regular bart rider, I assure you are plentiful).
You could shift the model around and maybe have a “station attendant” rather than a “driver,” but it’s an issue that I don’t see addressed in many autonomous conversations. The bus driver doesn’t just drive the bus, they also regulate and deal with all the bullshit happening on and around the bus.
Automation-in-public has so many issues beyond it’s closed-course equivalent, where you simply have to drive the vehicle. It will be a much bigger adjustment than just high quality autopilot.
i cant wait to see the testing videos here. think Boston Dynamics kicking their machine and pushing them downstairs. There better be billions of dollars in destroyed airframes ahead of any human cargo.
I would suggest building a tiny little system that is actually safety certified as per DAL DO-178B and can run on an aviation grade platform.
This suggestion applies to all "autonomous vehicle" startups. Build a small component that follows safety standards and is portable to major safety-certified platforms - this will be useful.
The military has said this is going to be a big thing.
I can’t remember which general it was, but during testimony in front the Armed Service Commission in Congress either this year or late 2017, he made the remark that around 60% of casualties in both Iraq and Afghanistan were essentially “logistics casualties” where people were injured or killed transporting something.
This technology will obviously reduce those kinds of casualties significantly.
On the other hand, warfare used to be limited by the size of somewhat small armies. Then came mass conscription, where the whole nation was drafted.
Then came “total war” where 16-65 year old men and women were fighting, but even then not everyone could fight due to someone having to support the frontline troops.
Perhaps this all ushers in another step of a more bloody, more people fighting each other, stage of war where everyone can be on on the “front lines”. The advent of cyber conflict, which extends the lines of warfare beyond what even strategic bombing could do, suggests this.
EDIT: Also worth noting that in air war, the bottleneck has usually not been plane production so much as it’s replacing the pilots. Air campaigns were limited by the amount of skilled pilots you were willing to lose. This also might turn that logic upside down as well.
Law enforcement & traffic reporting with AH,
a natural early application.
AH taxi services, sure eventually.
AH for LEO or EMS response to accident scenes
or other helo-accessible locations, sure.
But AH to transport emergent patients to
definitive care sounds like a marketing,
regulatory, and training nightmare. Fifty sets
of state laws, thousands of hospital/EMS medical
directors to work with, and tens of thousands
of local EMS workers to train seems like a long
hill to climb. Granted, we have had decades
of helicopter usage in EMS, so a path is known.
My guesses are that the “air ambulance” image is
being used to attract eyeballs and investors,
that early use cases will be for law enforcement,
with a pivot to expensive taxi/commuting usage.
22 comments
[ 3.0 ms ] story [ 46.1 ms ] threadI believe the tech can be made to be safe and reliable during normal flight. However, all the edge cases here are potentially fatal.
How well can we train a computer to choose a safe emergency landing spot and auto-rotate down if the engine fails? How will it cope with avionics/electrical failures? Or a geomagnetic storm?
How well will the system be hardened against malicious attacks? How well can we endow it with the judgment to avoid icing at low temperatures?
The answer to most of these questions might well be “pretty darn well, actually;” however, emergencies happen, and in those situations, I— and, I suspect, many others— will only ever feel safe with an experienced, context-aware, human pilot at the controls.
(Unless we somehow crack general AI...)
You might ask where this data will be coming from. I suspect 3D physics simulations are accurate enough that they could be used to generate data. Particularly if optimized for the physics behind helicopter handling.
A strict utilitarian should only care about accident rates. But as Greene [0] and others show, most people don't make moral judgements in a strictly utilitarian way. Thus, we judge homicide as worse than choosing not to save someone from death, even though in both cases someone dies as a result of a decision.
One of the lessons from probing moral calculus is that nobody can really articulate reasons for their actual moral decisions. People can articulate utilitarianism, but not what people actually do.
But, if you can articulate why you might prefer a human pilot over an autopilot, even given sufficient statistics that the autopilot is safer, it'd be interesting to hear.
[0] http://www.joshua-greene.net/
[0] https://www.aea.net/events/rotorcraft/files/US_Rotorcraft_Ac...
One of the biggest factors is that it is terrifying to imagine falling out of the sky in an aircraft under the command of a software-based pilot with which I cannot communicate and that I cannot see doing everything it possibly can to avert disaster. It may not be a rational fear, but we humans are not rational creatures.
It’s an issue of trust, which the folks interviewed in the article do acknowledge, to their credit. We can instinctively judge whether or not to trust another human being, because we have a lifetime of practice interacting with them. The burden of proof of safety for an uncommunicative mechanical system is much higher, and lots of testing (that I can observe on video) will go some way towards to building that trust.
Furthermore, things like avoiding thunderstorms, the motion of which we still cannot predict with computers (turbulence being an unsolved problem in physics), could be very tricky for an automated system to handle.
As autonomous technology progresses and we accumulate data, I might begin to feel differently. But for right now, without knowing all of the details of the system, I have no way of knowing how it is reasoning and perceiving the situation (I’m very familiar with the human consciousness, as I possess one myself) and this leads to a fundamental sort of uneasiness in my lizard brain.
I’m still bullish on technology, though, and this is an area in which I would be enthusiastic to be proven wrong.
You could shift the model around and maybe have a “station attendant” rather than a “driver,” but it’s an issue that I don’t see addressed in many autonomous conversations. The bus driver doesn’t just drive the bus, they also regulate and deal with all the bullshit happening on and around the bus.
Automation-in-public has so many issues beyond it’s closed-course equivalent, where you simply have to drive the vehicle. It will be a much bigger adjustment than just high quality autopilot.
> A new flying ambulance service will use small helicopters outfitted with tech that could eventually let them fly without pilots.
The actual title and subtitle. They have pilots right now.
Tail gearbox failure https://www.youtube.com/watch?v=11TD0Dboixo
ground resonance https://www.youtube.com/watch?v=0FeXjhUEXlc
lightening strike https://www.youtube.com/watch?v=uybTDikWzWs
This suggestion applies to all "autonomous vehicle" startups. Build a small component that follows safety standards and is portable to major safety-certified platforms - this will be useful.
This seems to be solving the wrong problem...
I can’t remember which general it was, but during testimony in front the Armed Service Commission in Congress either this year or late 2017, he made the remark that around 60% of casualties in both Iraq and Afghanistan were essentially “logistics casualties” where people were injured or killed transporting something.
This technology will obviously reduce those kinds of casualties significantly.
On the other hand, warfare used to be limited by the size of somewhat small armies. Then came mass conscription, where the whole nation was drafted. Then came “total war” where 16-65 year old men and women were fighting, but even then not everyone could fight due to someone having to support the frontline troops.
Perhaps this all ushers in another step of a more bloody, more people fighting each other, stage of war where everyone can be on on the “front lines”. The advent of cyber conflict, which extends the lines of warfare beyond what even strategic bombing could do, suggests this.
EDIT: Also worth noting that in air war, the bottleneck has usually not been plane production so much as it’s replacing the pilots. Air campaigns were limited by the amount of skilled pilots you were willing to lose. This also might turn that logic upside down as well.
Law enforcement & traffic reporting with AH, a natural early application.
AH taxi services, sure eventually.
AH for LEO or EMS response to accident scenes or other helo-accessible locations, sure.
But AH to transport emergent patients to definitive care sounds like a marketing, regulatory, and training nightmare. Fifty sets of state laws, thousands of hospital/EMS medical directors to work with, and tens of thousands of local EMS workers to train seems like a long hill to climb. Granted, we have had decades of helicopter usage in EMS, so a path is known.
My guesses are that the “air ambulance” image is being used to attract eyeballs and investors, that early use cases will be for law enforcement, with a pivot to expensive taxi/commuting usage.