So not to be needlessly critical but this is not news. Of course a robot is faster in a known map with perfect state information. They have always been. The problem has always been exactly those two things: static map known beforehand, and perfect state information.
The paper also don’t claim this as the contribution so this article is just… misinformation? I think there’s a word for this, willfully being flabbergasted by basically anything so you can write an article about it.
When before have drone robot quadcopters been faster than the best human drone racers, do you have any examples of that - a source or link? Are you speculating about theory or talking about a real event that happened, aside from this one?
Maybe what you mean is that it’s not surprising, because it was inevitable. That I would completely agree with. But it’s simply not true to say that robots have always been faster than humans. There was a first automated quadcopter that beat skilled humans, and it happened recently, because quadcopters are a recent development, and automated quadcopters are even more recent.
They're not saying perfect drone pilots have always existed. They're saying that designing a perfect drone pilot is easy and not worth making a fuss about.
I'm honestly very surprised that this is the first time it's happened. I would have thought it would have been done years ago. But probably there's not been enough interest to actually do it.
For anyone who finds it as worthless as I do, the original press release the article mangles is at [0] and the DOI of the paper is 10.1126/scirobotics.abh1221.
As for the work, it's one thing to say "Of course a robot is faster in a known map with perfect state information.", it's another thing to actually _build_ a working system.
Research is an incremental process and this seems to be like a meaningful step.
The new part here is that they have found a way to find the optimal path without using simplifications. So in a way, the true progress of the paper is a new mathematical loss minimization technique.
Just because you have all the information doesn't mean you can solve a constraint system before the heat death of the universe. Otherwise, NP hard problems like traveling salesman wouldn't be so scary.
Yea I think this is a bit harsh. I don't think it is quite such a trivial task to figure out when a FPV drone has such degrees of freedom. It can basically change to any direction at any time. Then the known space is the air in the room.The humans are also training on a known course.
A legit drone racing pilot is incredible at this also so it is not like there is a ton of meat on the bone to pick at.
It is cool from the perspective of racing drones even if less impressive from the perspective of AGI or something.
A drone can in fact not accelerate in any direction at any time. It can only accelerate along the thrust vector which is the normal of the plane that the rotors sit on.
That’s a distinction without a difference. In theory, of course you’re correct. In practice, your parent comment is correct.
The rotation rate in the roll or pitch axis is around 1080 degrees per second - 3 complete revolutions per second. Many freestyle pilots fly higher rates than me.
I can, and do, go from 80mph in one direction, flip 180degrees to accelerate back to 80mph in the direction i just came - a 160mph change of speed in around 5-6 seconds approx.
The only axis i cant turn very fast in is yaw (quads have poor yaw authority compared to other axes) but even then it’s fast enough most people would consider it instant.
One thing to keep in mind is that often times those products are serving a real need for niche communities (such as those with disabilities), and marketing them more broadly is simply a way to recoup the costs.
> Of course a robot is faster in a known map with perfect state information. They have always been.
Huhwut?
That would be news to an entire field of engineering, thanks. Please apply for your PhD.
Excluding the fact that whole classes of problems are NP-complete and difficult to compute, there are classes of problems like "How do I fit this odd shape through this odd obstruction?" that don't even lend themselves to being computed well.
I love the ETHZ has this department of playing with quadcopters. They may have shut down the Flying Machine Arena, but clearly the main theme survives.
This paper is not about machine learning. "Training" has nothing to do with the approach. External cameras are used because this paper is about trajectory generation and not about vision.
The paper presents an approach of generating a time-optimal trajectory through waypoints given physical limitations of the underactuated system. This is interesting and novel, and as demonstrated works very well. The group from which they come also work a lot with high-speed machine vision, and one of the next research steps will be combining this trajectory generation algorithm with onboard computer vision.
I wonder if we can take hundreds or thousands of such 3D map and accelerometer log pairs in order to train a model to be able to understand how to generically approach any new course.
The future could be humans inventing and learning new skills, then an algorithm quickly mastering and automating them (with the help of those humans)… makes you wonder if that kind of neural plasticity is even possible if your job was invented and made obsolete even faster than today. Can humans get better at this kind of creative flexibility?
On page 32 of the paper they provide rankings of the two pilots against whom the automated system flew (Michael Isler and Timothy Trowbridge). My personal opinion of course, but I would call both of these pilots "serious human competitors" given both have been competing since 2017 in international events and received many podium finishes. But that's also beside the point.
This paper is about generating time-optimal trajectories through waypoints given the system's physical constraints (e.g. limitations in thrust and rotational rates). A time-optimal trajectory is a trajectory which is time optimal—meaning that no faster trajectory exists. Given that this algorithm generates the fastest possible trajectory through the waypoints given the physical constraints of the system, it would be impossible for even a "serious" human competitor to beat it.
Perhaps, this could be used in simulation training to bring up the level of the top pilots. It could illuminate where they are losing time to an ideal pass.
So obviously the computer vision part here is nothing novel, but also the algorithm itself seems like the kind of problem that computer game developers have likely solved a few hundred times already, thought nothing of it, and moved on.
You are incorrect. Generating trajectories is easy. Many well known techniques exist that do pretty well, and yes this is done in computer games all the time (as well as in many other fields).
Quickly generating time-optimal trajectories for under-actuated mechanical systems with actuator constraints is interesting, and as a researcher in this field I can assure you that the technique in this paper is novel and is interesting—if it were not it wouldn't have been published in the journal Science...
The next big research question: Will it be feasible to have humans in the loop for military operations or will all the killing and destruction need to be 100% controlled by algorithms?
Why still the fear and paranoia over drones? Especially small quadcopters?
If you want to be scared of military tech, be scared of cruise missiles and ICBMs. Killing from huge distances away at the press of a button isn't new.
The paper also don’t claim this as the contribution so this article is just… misinformation? I think there’s a word for this, willfully being flabbergasted by basically anything so you can write an article about it.