MIT Tailsitter Drone Acrobatics | Hacker News Reader

MIT Tailsitter Drone Acrobatics

aera.mit.edu

149 pointsnadavision50 comments

Interesting research. Assuming that the benefit of fixed-wing aircraft is efficiency in relatively straight flight, I'm wondering how this aircraft compares in efficiency vs agility against a quadcopter of similar weight.

What would be your guess? I'm reading this as describing straight line wing flight for long distances rapidly shifting to erratic motions if someone doesn't like it existing. I imagine it's much more efficient at that, assuming that shell is mostly batteries anyway.

Very neat to watch, I'd be curious to hear under what conditions it's worse than a quadcopter, perhaps stability or how quick it can turn.

We're thinking along the same lines, I guess I just expected the upsides and downsides of the approach to be quantified somewhere in the research as context for why it's meaningful.

> Very neat to watch, I'd be curious to hear under what conditions it's worse than a quadcopter, perhaps stability or how quick it can turn.

Well, a quadcopter can fly with one busted motor/prop, this thing probably can't.

Winged flight is definitely for superior distance and loiter times.

Good point on the redundancy.

Another obvious drawback would be more surface to get buffeted by wind when in hover.

There could be price savings too, right? Two motors less and simpler ESC gotta save something.

Did no one notice the balls? Those are there so a set of external cameras [think motion capture] can feed in an absolute, essentially perfect state estimate to the control algorithm. It's probably a run of the mill system from Vicon [company].

Essentially if you're running vicon you can make flying things do things like you could make them do programming them in Blender or similar, subject to the [pretty minimal to the human eye] time constants of the mechanical systems. Brushless speed controllers are pretty fast, servos are as well [but way slower]. The end-to-end control loops we are talking about are in the ballpark of 1khz easy and have been for quite some years.

If they had balls they'd take off the balls ;) Other than that it is essentially CGI in real life ;)

Sorry don't mean to be negative it looks cool guys. Now go make it actually cool.

I ain't got no darn PhD in control theory from some fancy skool er nuthin but my gut tells me for this situation it's the state estimation that actually composes the beavers tail under the wattuh of dis dat der prollem.

I certainly noticed the marker balls, but...

On a quick glance, they do not mention if they are using the external cameras for the control algorithm or just verifying the results. The QR code-like markers on the gates suggests that there is also some onboard cameras.

The statistics on the measurement errors suggests that they have a ground truth (from external cameras?) which they compare to some other source of measurement.

So I would not draw the conclusion that the tracker balls and external cameras are doing all the heavy lifting here.

> The QR code-like markers on the gates suggests that there is also some onboard cameras.

I would be carefull with conclusions like that. These facilities are usually shared between a lot of different experiments through the years. The presence of QR codes on the gates certainly implies that someone at least once thought they might want to use onboard cameras in some experiment.

Are they used in this project? You can’t really tell by just looking at the presence of the QR code.

Same as I can’t tell if you are hungry or not by observing the presence of an oven in your kitchen.

I'm pretty sure that the "facility" is a gymnasium with cloth on the walls and floor to protect them. And, the hanging gates are rigid pink insulating foam sheets, probably made and hung custom for this project.

> I'm pretty sure that the "facility" is a gymnasium with cloth on the walls and floor to protect them.

Plus the tracking system. Which is the big differentiator of course. And the fact that it is not used for basketball games but to test indoor drones.

What else do you think you need for an indoor drone testing facility?

> And, the hanging gates are rigid pink insulating foam sheets

Yes.

> probably made and hung custom for this project.

No.

Here is an earlier MIT aeroastro project using the same gates from 2021: https://aeroastro.mit.edu/news-impact/system-trains-drones-t...

And this is the point I am making. There are many student studying aeroastro at MIT. Not all of their projects are about small drones but many are. And if the small drone test they want to do fits into this room they seem to prefer it. And if one project makes some gizmo (like those gates) for themselves and it looks usefull they won’t throw it away, but chuck it somewhere for storage and then the future projects, such as this one we are just discussing, reuses them.

> Essentially if you're running vicon you can make flying things do things like you could make them do programming them in Blender or similar

It is not quite that easy though. Yes having an external tracker will give you a reliable, high quality source of position and orientation.

But that does not mean that you can “program them in Blender”. You still need to figure out what kind of trajectories your system can or can’t follow and how to map from position and orientation errors to actuation outputs.

If you can’t control the robot with external tracking then you have no hope of controlling it with internal state estimation, so if you already have the test facilities it makes sense to start with that. That way you are not debuging two crappy subsystems depending on each other and failing spectacularly.

Model airplane servos aren't "way slower" than brushless motors (with their propellers). Here's a higher-end servo that moves 60° in 68ms:

https://www.horizonhobby.com/product/a7090-brushless-low-pro...

I doubt a motor-prop pair could significantly change its speed in 68ms.

Seeing it fly side-ways in circles (2 minutes into the 1st of the final 3 videos) is pretty wild. Not sure how the 2 rotors can create sideways thrust...

A slight tilt in the direction of flight?

Surprisingly it's going the same speed of 8m/s as the previous forward flight...

I don't think anything implied that was the top speed, more that the system can regulate its speed to match what was requested.

The top speed for winged flight is likely higher than helicopter mode.

You can tell by the sound that the motors work much harder in that mode. It could probably go 3x fast in normal mode.

A combination of differential throttling and the flaps. The flaps are in the prop wash so they don't need airspeed to exert forces.

Flatness[1] based control is pretty neat.

Basically, you can define a trajectory (that is differentiable n-times) and then calculate the state of the system from that.

In this case, given the trajectory they could compute the speed, acceleration, jerk and yaw+rate for the tailsitter ahead of time using the model.

[1] https://en.wikipedia.org/wiki/Flatness_(systems_theory)

>differential flatness

I am trying to understand this a bit better. Do you have other examples?

"Differential flatness" is synonymous to "flatness".

The concept is a bit comparable to inverse kinematics. With inverse kinematics you compute, e.g. the joint angles of a robot arm given its endeffector pose.

While with differential flatness you use the pose trajectory and its derivatives (i.e. how the pose changes over time) to calculate the state trajectory (joint angles, speeds and accelerations for the robot arm example).

Do these designs scale up? Don’t see why we haven’t created full sized aircraft with this form factor. Seems like it would make a great aerial weapons platform.

I'm pretty sure many of the new military designs (f-117) are inherently unstable without a machine to constantly adapt their control surfaces. this is just going all in.

Fun fact: the F-117 had its first flight in 1981 and has already been retired from combat duty for nearly 15 years[0]. So “new” is perhaps not quite the right word to describe it.

[0] https://en.m.wikipedia.org/wiki/Lockheed_F-117_Nighthawk

It’s definitely been done before.

The closest to this design off the top of my head would be the Lockhead XFV (1954): https://en.wikipedia.org/wiki/Lockheed_XFV

The idea of a VTOL tailsitter, but without the single primary planar wing surface, goes much further back. The Nazis were in the process of building the Triebflügel at the end of WW2: https://en.wikipedia.org/wiki/Focke-Wulf_Triebflügel

yeah there were a handful of tailsitter designs that era. Cool concepts but pretty much all had an achille's heel or two.

Having a human in the controls, trying to land it without seeing the ground, was a major hindrance. Drones don't really care about that.

Yes, though practicality to date has been limited.

<https://en.wikipedia.org/wiki/Tail-sitter>

It's all fun and games until the engine or propeller goes out....

Mustard on YouTube did a video on the French SNECMA Coléoptère, which had a ring for a wing. It tended to rotate while hovering, and hard to steer during landing.

(9 min) https://youtu.be/unz6mfjS4ws

Interesting, thanks. At the end (~6:50) there are a few seconds about the aerodynamics of the 'wing', which had been my first question.

Probably nowhere near this performance. Their motors instantly vary thrust by a factor of ten or more, this just can't be done at human scale.

Did you forget about adjustable pitch propellers? We have managed to use human-scale ones even to synthesize low-frequency audio by building a sealed room and putting a reverse-thrust-capable one on the only opening of that room (besides a closed door). It spins and the pitch is adjusted to modulate thrust in response to the sound signal.

Those rotary woofers deliver supposedly good distortion up to their rotation frequency; 13 Hz / 800rpm seems common. For reference, a tail sitter propeller would probably spin faster than that, though even this is plenty to control the aircraft. A blade design using a servo flap easily keeps up with those control speeds, and could likely even offer thrust vectoring like the rotor of a helicopter (could eliminate the flaperons).

The real enemy is power-to-weight ratio.

Ah, tailsitters. Everyone loves the theory, then the practical realities of the takeoff position exposing the entire wing surface to the prevailing wind quickly kills real-world applications. I remember back during the 3D Robotics heyday watching Chris Anderson repeatedly run down to set his tailsitter upright, take a few steps back, only to watch it fall over again.

For vertical takeoff and landing the wing surface is mostly irrelevant, right? Seems like if you gave the rotors a “landing configuration” where they rotated 90 degrees you could lay the wing flat on the ground.

The wings look like they have smaller control surfaces the full wings are not. But to make cheap drones that are dropped or launches from a cheap taco holder its probably fine.

This is the distinction between tilt wing aircraft and tilt rotors. The former were explored in the 60s and found impractical. The latter took longer to develop but are now practical.

The original comment is spot on: hovering vertically with a big wing also vertical just doesn't work well in the real world. That said, the situation might be different for very small drones, for similar reasons to why you don't build full size quadcopters.

Movable motors would greatly increase the complexity and weight.

The Harrier, Osprey and the carrier variant of F-35 are real world aircraft that point the thrust vector at the ground for takeoff and landing, then rotate it to the back for forward flight.

Can't you take off from a closed box? For landing the flight controller should be able to compensate for the wind and come down in a somewhat vertical position.

There are commercially available tailsitter UAVs primarily used for mapping purposes:

https://www.atmosuav.com/product/mapping-drone-marlyn?view=w...

It transitions into forward wing flight after take off. It does achieve a greater flight distance/flight time per battery compared to the same size of thing as a quadcopter, but NOT as great of endurance as something like a 2 meter wingspan VTOL with four lift motors + single thrust motor, such as:

https://www.foxtechfpv.com/foxtech-loong-2160-vtol-mapping-p...

I (as someone without a need for any drone and therefore without having looked at enough prices of drones to have any gut instinct here) was curious how much that first one you linked - the Marlyn UAV - might cost, since the company only says to contact them for pricing.

This page [1] suggests $17k, is that likely on the money (pun intended, sorry) or just a random scam site?

[1] https://volatusdrones.com/products/atmos-marlyn

don't think that's a scam site, though it probably has a healthy profit margin. that price is within the range of normal for a fully built surveying drone for use by people who aren't interested in getting into the technical details of pixhawk/ardupilot/arducopter stuff and building their own. They're sold to surveying companies, mining companies etc.

go look at pricing for the DJI Matrice 30 (M30) for some comparisons...

Martin UAV is looking like the one to beat in medium range right now, and it's a pretty slick tailsitter design. If I was a betting man, I'd say that it's going to eat ScanEagle's lunch sometime in the next few years, or get bought. Which, ugh, the latter seems more likely, sad to say.

I'm not with Martin, but scuttlebutt is that the flight controls take into account takeoff and landing winds, then use that to adjust the flight attitude on approach and takeoff, since the wind can add to the effective airspeed for rotation.

Once it tags the ground the flight procedures have it nail itself to the pad double quick. That's about the only dodgy part, but they've done it from the back of a speeding truck and it looked fine. Hell of a lot finer than "drive the plane into a rubber band hanging from a stick".

I'm curious if it'd be harder for a soldier to spot a tailsitting drone high in the sky floating above doing surveillance.

They are already impossible to spot as it is at certain heights.

It would have to be right above them though.

Does visibility matter?

I imagine a drone hunter flight of multiple drones, multiple drones implies multiple receivers, multilaterating the transmissions of the target, and splash.

Only works on surveillence drones that need to be transmitting, and don't emit only in the direction of the intended receiver.

Based on what we're seeing in footage and interviews from Ukraine it absolutely matters. Hearing a drone is often the warning that you're about to get shelled.

People really fixate on this swarming concept like that murderbots video and breeze right past the more high impact basics.

Good point that a vertical wing is hard to see from below, but quad copters don't have wings, and so don't have that problem (the rest of the body has width, and is visible, but that's mostly the same with a tailsitter).

Yeah soldiers in Ukraine said that the Mavic drones are basically invisible to people on the ground. Unless they spend a lot of time looking for them.

It's probably not the biggest issue.

One thing is whether they can aim a grenade dropping drone easier or would the wind shift it even more.

With a completely static environment the drones will perfectly exactly to simulation, within the tolerances of their hardware. Try it outdoors.

Reply on news.ycombinator.com