3 ms·

It is fascinating that they solve robustly 50 times per second a control problem and handle gracefully numerical issues. Granted they have dedicated hardware to

It is fascinating that they solve robustly 50 times per second a control problem and handle gracefully numerical issues. Granted they have dedicated hardware to do so that fast, but it never stops to amaze me.

tonyarkles 5y ago
In some ways, a quadrotor's control system is a more challenging controls problem! And I'm not at all suggesting that the rocket control loop isn't hard, but a few things to consider: - Quadrotors control their attitude solely by adjusting the thrust generated by the four motors. Unlike rockets, or fixed-wing aircraft, or cars even, the moment there is any kind of disturbance (whether software, hardware, or environmental), the aircraft immediately starts to fall out of the sky. - For the work I do, we typically fly around 5-7m above ground. If the motors completely quit, we have around 1.2 sec before we hit the ground at 42km/h. - The aircraft is also capable of roll rates >50deg/sec. Even if you're high enough that you don't immediately hit the ground, motor problems can also result in you pointing upside down and pulling yourself towards the Earth instead of pushing away. - Most of the parts have notoriously bad repeatability, plus small variations would wreck any repeatability anyway. You can't just set all of your motor throttles to 60% and expect them to have the same thrust, it's nested-upon-nested control loops to keep everything going - The sensors are generally crap too with non-trivial amounts of bias and drift. And the wildest part of all of this? You can get a top notch flight controller to handle all of this for you (and open source!) for $150 USD https://shop.holybro.com/pixhawk-5x_p1279.html https://shop.holybro.com/pixhawk-5x_p1279.html?
PaulHoule 5y ago
The data rate for flight control systems (air or space) is not that great but the requirement for reliability is. That’s not just an absence of software defects but also reliability in the face of hardware failures. (Compare the A320 flight envelope protection to the 737 MAX MCAS.)

tonyarkles 5y ago

In some ways, a quadrotor's control system is a more challenging controls problem! And I'm not at all suggesting that the rocket control loop isn't hard, but a few things to consider: - Quadrotors control their attitude solely by adjusting the thrust generated by the four motors. Unlike rockets, or fixed-wing aircraft, or cars even, the moment there is any kind of disturbance (whether software, hardware, or environmental), the aircraft immediately starts to fall out of the sky. - For the work I do, we typically fly around 5-7m above ground. If the motors completely quit, we have around 1.2 sec before we hit the ground at 42km/h. - The aircraft is also capable of roll rates >50deg/sec. Even if you're high enough that you don't immediately hit the ground, motor problems can also result in you pointing upside down and pulling yourself towards the Earth instead of pushing away. - Most of the parts have notoriously bad repeatability, plus small variations would wreck any repeatability anyway. You can't just set all of your motor throttles to 60% and expect them to have the same thrust, it's nested-upon-nested control loops to keep everything going - The sensors are generally crap too with non-trivial amounts of bias and drift. And the wildest part of all of this? You can get a top notch flight controller to handle all of this for you (and open source!) for $150 USD https://shop.holybro.com/pixhawk-5x_p1279.html https://shop.holybro.com/pixhawk-5x_p1279.html?

PaulHoule 5y ago

The data rate for flight control systems (air or space) is not that great but the requirement for reliability is. That’s not just an absence of software defects but also reliability in the face of hardware failures. (Compare the A320 flight envelope protection to the 737 MAX MCAS.)