Military robot "lands" B737
 

I love how to takes so long to apply reverse thrust and still has it selected as it comes to a complete stop.

I won't be worried just yet.

Maybe his robot friends will repair the engines?

A demonstration of how not to automate a system. Rule 1 is "Automate the function not the procedure".

I wonder if they've perfected the external inspection yet. :8

Can someone tell the management?

From what I can see, the aircraft actually did an autoland, and all the magic robot did was to select reverse. No hands of feet keeping it straight on rollout, and certainly not a single-engine manual landing in a howling crosswind. The job's safe for a while, Guys.

The company behind this (Aurora) have developed several fully automated (pilotless) aircraft before, including the Centaur:

https://www.youtube.com/watch?v=a13ytpsQ_Xs

I think in the 737 sim they were demonstrating the ability for a robot co-pilot to work with a regular pilot (human Captain) as part of an augmented crew. The robot (ALIAS) is able to listen and understand the Captain's commands (it also has speech synthesis to respond). The robot is also able to take over flight duties and land the aircraft in case the Captain becomes incapacitated.

ALIAS stands for "Aircrew Labour In-Cockpit Automation System".

An autoland followed by some very poor lever actuation that would have to be repeated in a sim detail. Nothing to worry about.

Problem is, it has to go into the hold during flights to USA

Love it!

Not if they can get it down to the size of a mobile phone. But then it couldn't reach the controls.

don't worry, with the race to the bottom in pilot salaries, it will be a long time before this kind of automation becomes viable.

Surely they can automated the application of thrust reversers without having to have a robotic arm mechanically do it?

That was surprising to me as well. One would think advances in automation would go in the FBW way (i.e. from computer directly to the actuators). I don't see much point of building robots to fly the airplane through the interface designed for human.

The number of commercial & military aircraft today which cannot (and will not) be retrofitted with new automation is around 100,000. Many of these aircraft will still be in service 20 years from now. Augmenting such aircraft with a robot without requiring any modification / recertification isn't a bad idea.

However, this robot in particular is simply a technology demonstrator, i.e., that robots can collaboratively work with humans using "natural" interfaces (voice, vision, controls).

It would have been easily to program the robot (or autopilot) to automatically apply the thrust reversers, but that wasn't the point. ALIAS in this case applied the thrust reversers without prior programming -- but instead by listening to the Captain's voice commands and executing them.

The future use for robots like ALIAS may or may not be in the cockpit at all, but perhaps in assistive fields (medicine, home/elderly care, construction) besides various military applications.

I guess they could have put the 'robot' in any sort of environment to test it though the quality of feedback they need to prove and improve the product would be lacking. Airline simulators are probably the most demanding human/machine workplace interface environments to be found, and they come with people who's job it is to identify and access a pilots skills ...or robot skills. For the cost of hiring an airline sim for a day or two it is probably far cheaper then setting up a dedicated test facility

I like the accuracy of Sky reporting (as always) 'essentially seeing things just as a human would' Eeer, No. Besides, it wouldn't be able to see much as, during the roll out, it had it's head down looking for the thrust levers - needed to keep looking out the windows too. :}

Off to JB with this thread.

Dave Parnas


A good introduction to decision making limitations & issues predominately related to AI, but also applicable to automation is "Rationality: From AI to Zombies" by Eliezer Yudkowsky, a reformed AI scientist. Its available on a "Pay What you want" (read $0 if you want) basis HERE.

He also has an excellent blog LessWrong.com on cognitive biases. Lots of good info for pilots on how our own human decision making is flawed, but in predictable ways.

it seems totally pointless to me - given that majority of the controls manipulate themselves when on autopilot/autoland, why not just automate the whole cockpit and 'cut' the robot out? if you want the automation to 'work' with the pilot put in AI interface with voice recognition....still think that the overall concept of removing humans will eventually be realised anyhoo

Again, automatic the cockpit just means a smarter autopilot, and requires extensive retrofit which may not be possible for all planes.

But a collaborative robot can be deployed everywhere... not just in the cockpit (even existing ones), but in the hospital, home care, military, construction, factory, etc., without the need to retrofit anything.

I doubt this robot can be "deployed everywhere" without job specific customization of its extremities and tools. What works in a cockpit won't work in a hospital, and the hospital robot won't be suitable for construction work.

Despite your claim to the contrary, the robot that "landed" a 737 had to be programmed for the task, including the XYZ coordinates of every device the robot touched and operated. How will a robot be useful for home care unless it knows everything about its surroundings?

I suppose the answer will be "AI". But that sort of thing is decades away from becoming ubiquitous. I constantly see claims that self driving cars will become a reality in less than ten years. I don't believe it.

Current self driving automobile prototypes have sensors and computing power that cost well over $100,000. The public won't be willing to pay a huge premium for a self driving car, so there's quite a bit of work that needs to occur before there is one in your driveway.

You're still not seeing the big picture. The point isn't to "avoid" programming. The point is to create a general purpose system so improvements in one area can be applied to other areas.

I'll give you two analogies:

1. The very first computers were task specific. There was one computer just to compute oceanic tides. Another computer to do statistical processing. Yet another one to calculate trajectory of artillery shells. These computers were created for specific "verticals".

Then came general purpose computers. Instead of just improving task specific computers, now technological advances (e.g., in memory, storage, computing power, etc.) can be shared among all possible programs, then and in the future.

This change from "vertical" to "horizontal" thinking revolutionized the computer industry and literally changed the world.

2. Look at your smartphone today. Think of how many functions your smartphone can do today: it's a phone, a camera, a modem, a bar code scanner, a music player, a stopwatch, a metronome, a pedometer, an alarm clock, a flashlight, etc.

All in one general purpose device! Remember not so long ago, each and every one of these functions were all separate task-specifc hardware.

Personally I'm glad we didn't simply "improve the stopwatch" or build a better flashlight, the way some here are suggesting that we should simply improve the AP.

There's value in creating general purpose systems instead of always thinking in specific verticals. The ALIAS robot is one step in that direction.