Humans, Robots, and Non-Prehensile Manipulation 

I competed as part of a team in the DARPA Robotics Challenge Finals; crazy times. Now I'm watching this because I'm finally at the point where I'm starting to learn about how in the heck our team avoided falling. Thank you so much for your videos and the work you've done to keep education accessible! This is priceless.

Woah!! This video so much inspired me. I thought it `s nothing when I saw the butterfly robot motion and i didn`t expect that it d be that much hard for human. now I understand why that`s revolutionary technolgy. Thank you for share this video!

A colleague of mine from Mexico has pointed out the comment from Pedro, and I think I am in the position to clarify since both Pedro La Hera and Daniel Morales were my PhD students (and MSc students before that), and Maksim Surov was a postdoc supervised by me approximately the same period in Umea, Sweden. That time Pedro and Daniel primarily helped me with developing experiments and implementations with new methods for controlling hydraulic cranes for an industrial partner. Meanwhile, I have been working on creating new control algorithm for underactuated systems for more than decade. I have shared and taught graduate students and Pedro and Daniel, as others, had studied the topic, see the link for the (not updated) course folk.ntnu.no/shiriaev/files/oscillations_in_mechanical_systems.php. The butterfly example has come into my view after discussion with Dr. Kevin Lynch in 2009 and Dr. Giusseppe Oriolo who suggested it as a challenge for our theory. I have accepted the challenge and discussed and mentioned the problem with many students. The problem/challenge with new example come from the presence of unilateral constraint absent in the previous examples and absent in the general approach proposed by me and colleagues (Leonid Freidovich and Sergey Gusev). I had mentioned that to Pedro and Daniel as well, however, they had overlooked the point and produced a replica of my general arguments (previously iterated for many other robotic systems) for this example. It is perfect when students comprehend the difficult theoretical material, used it and write the text. At this point the fact that this technical exercise does not solve the original problem is not important for them and for the text. But it is important for the Butterfly robot example! I should say that the progress has come with specific parametrization of rolling motions proposed by Dr. Maksim Surov, which substantionally advanced the known to me arguments. Maksim has afterwards spent two year for implementation of this theoretical contribution and built three different platforms to validate the result. Of course, if some of arguments of Daniel and Pedro would be used in this development, then they should be mentioned in the paper in the list of coauthors. However, this was not the case, and the comment of Pedro indicates that they should be more critical for themselves as scientists and fair in judgments. Several leading US and world-wide researchers have tried to find the way to solve that example for decades, our team was the first one who made that beautiful show. All the participants who contributed to the result are listed as authors of the contribution and recognized.

The hand built version has an axis tilt that is not there in the original problem ... therefore it is a much much harder balancing problem. A fixed shaft that can only rotate seems in the frequency range people would be able to cope with. The bigger butterfly robot actually has a fixed horizontal shaft.

the interesting part is the fact that this area needs a lot of research and work, so everyone has a chance to leave his fingerprint. And you're uncontrollably awesome, thanx for the videos ;)

Thank you , you are the best instructor in the field of control. Please don't stop providing content.

I have two questions related to a different topic and I would really appreciate it if you answered them :) First question: Is there any application/s where I need the PID values of a controller in a system to be frequently tuned? I can think of some applications but I don't know whether I should change my PID values or not! These applications are 1. In humanoid robots where the environment is dynamic. 2. Another application might be when in quadcopters where the weather/altitude suddenly changes or when a single motor malfunctions so the other motors should be re-tuned. The second question is: If i gave a system a parabolic input to follow, should i feed my controller with only one PID value or several PID values? where each PID Value will be associated with each different slope! Thanks a lot :)

You're awesome, Brian, keep up the great work!

Great Video Brian. I guess I've seen this video 6 times already and I'm still coming back. Way to go Pal!

I enjoy to see how some people are talented at explaining complex problems of nonlinear control in simple ways. However, as a scientist, I think that it is always important to keep facts, particularly in public material, as it is the case of youtube. In fact, the problem of controlling underactuated systems, such as the butterfly robot, balancing pendulums and so forth, are quite complex in theory. And what I think is important to point out is that the theory required to perform motions with a setup like the butterfly robot has in fact been solved much earlier and by other people to what it is mentioned in this video. The only contribution behind the video of the butterfly is the practical implementation, but the scientific development, theory, and all work prior the experimental tests have been solved by others. And for those interested on such theory, all of this is clearly explained in the original work described in the PhD thesis of Daniel Ortiz Morales, and the simulation software and hardware design can be asked to him by email. With 3D printers widely available, it is simple to replicate these solutions with just a webcam, some motors and a simple processor like arduino. This is just a comment.

Informative and entertaining. I have been in ICRA this year but IDK how I missed the butterfly shaped robot.

Crazy how this was just 6 years ago and now we have Boston Dynamics Atlas running, doing flips and jumping. And spot.

i wish you would upload more, i watched all of your control theory vids

This is a very interesting topic. May I suggest a different approach? Something that in my opnion is a really hot topic now In robotics. You are viewing this purerly as a control problem. However for the human part of the probem this is actually a Machine Learning Problem. The human brain can be subtracted with a Neural Netwrok and it is true that many control mechanism (such as a PID Controller) can be substituted with a neural network which acts as an interpolator. However, the problem is that the Neural Network must first be trained, much like a human brain first has to be trained. There is a lot of depth to this approach and it is something, which I personally believe is really hot in the field of robotics, but I cant summarize it here very well. To sum up, as a classical control theorist you are basically questioning whereas we can achieve stability in this particular case. However, modern machine learning approaches (Deep Learning, Neural Networks) actually state that with sufficient training a mechanism (such as a human brain) can DISCOVER the dynamics and controls of a particular problem. Very inspirational video indeed. Keep it up!

Awesome job, Brian, as usual. This one even piqued Erin's interest! Love the videos, keep em coming!

Great videos !! Appreciate the effort you put in !!

your videos are pure gold, I love it, I can procrastinate now by actually learning the stuff I'm procrastinating on.... You are a god among men.

I think one of the issue why you could not repeat the butterfly thing with your woody apparatus, because the axil of the rotation was not stable, wobbling. Also, the edges of the woody thing is not not smooth as the one in the show... I think one of the way the DARPA robots can be improves is by designing the legs of the robot as spider legs which puts the center of the mass as the lowest point. I guess if they implement some sort of MPC in them, which they might have done, it could improve the movement of the robot.

This was a cool video, I was expecting some maths though :P

Dear Brain, I love your videos and more importantly, your pation to controls! I am a control engineer persuing my Masters in measurements and controls and I just can't have enough of your videos! I teach a class on Mechanical controls and I keep refering them to your this channel. I just wanted to pass the love so you know how great of a job your doing. if I may ask, what do u do for living? much respect, AJ

MIT OCW has an online course that feels relevant to this that I wish I had the mathematical prerequisites to follow. It's called underactuated robotics and it's also archived at edx under MITx (and here on youtube, too).

Great video!

Waiting for more videos.. great job. :)

We actually are building a robot with point feet using the same class of algorithms This theory is developing for a long time but it is rather complex.

Hey Brian...great video...your videos on classical control theory are great, you helped me a lot understanding control theory. i have a question that Why do we use time domain analysis in real physical system..what REAL TIME stands for in a real life design like any modern medical equipment etc. reply plz

This video is so funny and I am enjoying it a lot! The big one, I never imagined that you will do that task with your so cute friends! These experiments give me a lot of energy and motivation! Thanks a lot to you and your friends!

Speaking of MPC, I would love if you make some video about MPC, thanks

how can someone know about these scientific words or nomenclature because i am actually doing M.Sc and one thing i find hard is to find journals because generally they all use these wording

loved it

Can you please do a video on Z transform ?

awesome...............

Nice vid Sir, may I share you video in my blog?

Are you going to make videos about digital control? We need you here in Brazil! haha

Maybe you've seen this by now, but MIT has been doing a ton of research into nonlinear robotics, and they've been creating a software package with their algorithms. drake.mit.edu/ I bet the butterfly robot would be trivial to do with their toolbox.

The robot design has already been made to solve your problem.

would one have more control over the large butterfly, if you control the rotation with something like a bicycle pedal?

u r awsome

Why not use full size Boba?

this video made me laugh so many times XD

Your Height?

hocammmm burası itü değil mi

So what I'm taking away from this is that short people require a higher bandwidth than tall people.

Basically, one step closer to creating a more stable and faster T800... Lets see if it's wise to invest into this kind of research. I'm just joking here...or am I? :)