Comment box Scope: observations Tone: intrigued, skeptical Opinion: yes Sarcasm/humor: none This sounds like a useful technology for reducing the cost of transporting cargo by air. It does also...
Comment box
Scope: observations
Tone: intrigued, skeptical
Opinion: yes
Sarcasm/humor: none
This sounds like a useful technology for reducing the cost of transporting cargo by air. It does also put some cargo pilots out of a job.
Machines tested to the extent the FAA demands typically make fewer mistakes than humans. They are great at following safety routines. From that perspective, this is also safer. Nice. I don’t know about the Air Force.
I am not a computer expert, but with that said, it seems careless to approve flight paths of any sort that could kill people on the ground in the event of a critical software error. In this case a human can’t take over. I imagine that most of the routes are rural, but I’m more concerned with takeoff and landing which will be in areas that have at least some people. Even a rural airport has occupants.
Computers are better than humans at doing mundane safety checks and following procedures. They are utterly terrible at improvising in an unusual situation, including fancy AI models. Not to mention that I doubt these tests have been conducted in difficult circumstances. What happens when a plane is flying through wildfire smoke?
2800 flight-miles seems low. The state of California is 760 miles long. That means these planes would have flown the equivalent of three and a half trips across the state. Realistically I imagine they were flying much shorter, set routes. That data could not possibly have represented a large number of geographies.
Given the fact that airplanes are largely unobstructed by natural features and can fly where they wish, fuel notwithstanding, it also seems careless to assume that a computer system will not (at some point) bug out in a way that sends it ridiculously off-track.
I suppose you have to start somewhere. I can’t say I’ve ever loved the concept of autonomous drones/planes though. We can’t even get transport automation right on the ground, except for some limited-access trains. I feel like running robot-operated aircraft is asking for trouble.
Based on a comment made at the bottom of the article and and company website, it sounds like there is a pilot “supervising” on the ground. I imagine it is similar technology to drones. They are...
I am not a computer expert, but with that said, it seems careless to approve flight paths of any sort that could kill people on the ground in the event of a critical software error. In this case a human can’t take over.
Based on a comment made at the bottom of the article and and company website, it sounds like there is a pilot “supervising” on the ground. I imagine it is similar technology to drones. They are mostly in autopilot but a human pilot can take control remotely.
Also, testing AIs in a computer is a totally valid way to do it. You can throw at it all of the insane situations you might encounter without having to wait for them, and it will respond exactly...
Also, testing AIs in a computer is a totally valid way to do it. You can throw at it all of the insane situations you might encounter without having to wait for them, and it will respond exactly the way it would in the real world. Plus all the interfaces between the computer and the airplane can be tested without actually flying--you don't need to be in the air to see if an aileron or engine is responding properly to control inputs. Yes, there are reasons that testing while flying can be advantageous, but you get almost everything you need without flying.
Plus, autopilot has been capable of making planes autonomously takeoff, fly to their destination, and landing for many years. Plus drones have been doing these things autonomously for years, and the military has had fully autonomous aircraft flying at least in tests for well over a decade now (possibly over two), including autonomous aircraft carrier takeoffs and landings. I won't say it's a solved problem, but the solution is trivial and left as an exercise for the reader.
Airplanes are easier in some ways because they can stay far away from everything most of the time, including each other and the ground. And there are no pedestrians. (Landing and moving on the...
Airplanes are easier in some ways because they can stay far away from everything most of the time, including each other and the ground. And there are no pedestrians. (Landing and moving on the ground are trickier.)
Comment box
This sounds like a useful technology for reducing the cost of transporting cargo by air. It does also put some cargo pilots out of a job.
Machines tested to the extent the FAA demands typically make fewer mistakes than humans. They are great at following safety routines. From that perspective, this is also safer. Nice. I don’t know about the Air Force.
I am not a computer expert, but with that said, it seems careless to approve flight paths of any sort that could kill people on the ground in the event of a critical software error. In this case a human can’t take over. I imagine that most of the routes are rural, but I’m more concerned with takeoff and landing which will be in areas that have at least some people. Even a rural airport has occupants.
Computers are better than humans at doing mundane safety checks and following procedures. They are utterly terrible at improvising in an unusual situation, including fancy AI models. Not to mention that I doubt these tests have been conducted in difficult circumstances. What happens when a plane is flying through wildfire smoke?
2800 flight-miles seems low. The state of California is 760 miles long. That means these planes would have flown the equivalent of three and a half trips across the state. Realistically I imagine they were flying much shorter, set routes. That data could not possibly have represented a large number of geographies.
Given the fact that airplanes are largely unobstructed by natural features and can fly where they wish, fuel notwithstanding, it also seems careless to assume that a computer system will not (at some point) bug out in a way that sends it ridiculously off-track.
I suppose you have to start somewhere. I can’t say I’ve ever loved the concept of autonomous drones/planes though. We can’t even get transport automation right on the ground, except for some limited-access trains. I feel like running robot-operated aircraft is asking for trouble.
Based on a comment made at the bottom of the article and and company website, it sounds like there is a pilot “supervising” on the ground. I imagine it is similar technology to drones. They are mostly in autopilot but a human pilot can take control remotely.
Also, testing AIs in a computer is a totally valid way to do it. You can throw at it all of the insane situations you might encounter without having to wait for them, and it will respond exactly the way it would in the real world. Plus all the interfaces between the computer and the airplane can be tested without actually flying--you don't need to be in the air to see if an aileron or engine is responding properly to control inputs. Yes, there are reasons that testing while flying can be advantageous, but you get almost everything you need without flying.
Plus, autopilot has been capable of making planes autonomously takeoff, fly to their destination, and landing for many years. Plus drones have been doing these things autonomously for years, and the military has had fully autonomous aircraft flying at least in tests for well over a decade now (possibly over two), including autonomous aircraft carrier takeoffs and landings. I won't say it's a solved problem, but the solution is trivial and left as an exercise for the reader.
Airplanes are easier in some ways because they can stay far away from everything most of the time, including each other and the ground. And there are no pedestrians. (Landing and moving on the ground are trickier.)