Safe2Ditch

[DefiantChild]

I ran into this Technology Review article this morning I thought I would share. In short, a husband/wife team have developed a solution to landing UAVs safely when battery life or other in flight impairments arise.


This software promises to teach drones how to crash safely in an emergency

 

[Goof]

"Once these remaining technologies are solved" (how to land the drone in the event the power is cut, or a motor fails)

Short of running Hexa or Octacopters and/or an entirely redundant set of motors, batteries, ESCs, IMUs and props, how does she expect software to magically save a dead in the sky drone? Deploy a set of wings and glide? Parachutes?

 

[DefiantChild]

I suspect they'll apply AI learning techniques to "anticipate" and act on impending motor & battery failures using a secondary autonomous monitoring system that does not rely on the primary AC control or power systems.

It occurs to me, in the rare occurance of a drone dying mid-flight, a fail-safe drag system (chute, find, prop) of some design might be mechanically deployed to control decent velocity within "safe" parameters. By safe, I mean safe when dropped on unsuspecting humans.

 

[jeffames226]

Ridiculous waste of time

 

[captainmilehigh]

Unless they find a way to defeat gravity, the most they can do is slow the descent. Either way, it's coming down....controlled or not. Far too many variables for this to work effectively. DC, thanks for sharing. I guess we'll just wait and see what they come up with.

 

[anotherlab]

I suspect they'll apply AI learning techniques to "anticipate" and act on impending motor & battery failures using a secondary autonomous monitoring system that does not rely on the primary AC control or power systems.

It occurs to me, in the rare occurance of a drone dying mid-flight, a fail-safe drag system (chute, find, prop) of some design might be mechanically deployed to control decent velocity within "safe" parameters. By safe, I mean safe when dropped on unsuspecting humans.

The second generation of Amazon drones are more like fixed wing aircraft. They should be able to glide if the motors fail and an emergency landing database would work for that. Assuming that the flight controls could still work. For a multi-rotor craft, by the time you would need Safe2Ditch, it's probably already game over.

 

[DefiantChild]

For a multi-rotor craft, by the time you would need Safe2Ditch, it's probably already game over.

Game over is more relative to altitude, rates of decent, and other factors including implementing proactive mitigation vs reactive. I believe the technology that the developers have in mind is proactive. In other words looking at and learning the behavior of sensors, other inputs, metrics, and flight paths to establish patterns or sequences that lead to failures and then proactively implementing a safe landing solution under similar circumstances.

In reactive, "oh ****" scenarios, the proposed tech may or may not apply. A "system" may not be able to "auto-react" in a scenario where a bird fails and falls from 25 ft (1.28 sec to impact). But conditions at 125 ft (2.8 sec to impact) could be an entirely different story and reduce force of impact sufficiently to avoid injury and/or damage.

You have to keep in mind that much of the push to impose potentially overbearing regulations on drone flight has to do with the perception and actuality of safety. Anything that "affordably" mitigates risk without sacrificing performance or imposition of regulations, benefits all of us.

For the record, I've done controls programming and am aware of what micro control circuits can do in fractions of seconds. So while it may be a complete waste of @jeffames226's time, it's obviously not a waste of others'.

Ymmv

 

[Goof]

This is interesting:

It's one thing to do this with an empty payload, for this to work with a laden drone the thing will need to be grossly overengineered.

See also:

One could deploy flaps under the arms to slow the yaw rate in the event a lift failure is detected. This could potentially allow for more control, although it looks like the current algorithm is doing OK by itself....

 

[Fester]

Game over is more relative to altitude, rates of decent, and other factors including implementing proactive mitigation vs reactive. I believe the technology that the developers have in mind is proactive. In other words looking at and learning the behavior of sensors, other inputs, metrics, and flight paths to establish patterns or sequences that lead to failures and then proactively implementing a safe landing solution under similar circumstances.

In reactive, "oh ****" scenarios, the proposed tech may or may not apply. A "system" may not be able to "auto-react" in a scenario where a bird fails and falls from 25 ft (1.28 sec to impact). But conditions at 125 ft (2.8 sec to impact) could be an entirely different story and reduce force of impact sufficiently to avoid injury and/or damage.

You have to keep in mind that much of the push to impose potentially overbearing regulations on drone flight has to do with the perception and actuality of safety. Anything that "affordably" mitigates risk without sacrificing performance or imposition of regulations, benefits all of us.

For the record, I've done controls programming and am aware of what micro control circuits can do in fractions of seconds. So while it may be a complete waste of @jeffames226's time, it's obviously not a waste of others'.

Ymmv

If it were not for flight control systems capable of perhaps millisecond corrections, no human pilot would be able to keep an F-117 or B2 in the air as they don't have much in the way of inherent stability. Don't know if current flight firmware could be updated to include this capability, but doubtless something similar will be required in the future.

 

[anotherlab]

...In other words looking at and learning the behavior of sensors, other inputs, metrics, and flight paths to establish patterns or sequences that lead to failures and then proactively implementing a safe landing solution under similar circumstances.

That sort of programming would be useful and identifying risky behavior could allow the control (DJI Go, Litchi, etc) to notify the pilot that the current flying could lead to AC failure.