As an embedded systems engineer who studied control theory at graduate school and had machine learning oriented internships, I find myself today in my career at the crossroad between all 3 fields. I just wanted to say that your work has been a major help in igniting my passion for control theory back in school and now I am finding out that you're bringing machine learning into it. I look up to you and am just fascinated by the work you do making this knowledge available publicly. I always entertain the idea of meeting you someday and picking your brain on various topics and engineering problems.
Just want to shout out how good this series is. I came away with understanding and inspiration. Anyone here at the start, settle in and enjoy, Steve makes this a joy to learn.
Found your series while looking for more information about data-driven modeling, machine learning and AI, as a software engineer interested in learning. The math initially made my head explode; it's been too many years since I reviewed algebra and worked through equations. I can feel my skull expanding as certain neural highways rebuild themselves while watching. Excellent lectures!
I found your video, because I searched for a solution to write a just in time interpreter for a just in time interpreter. Some kind of data driven subscript language on PHP. In the end it's a very high dynamic system. Your introduction brought me some vocabulary to think further. Thx a lot! 🤗
@Steve Brunton,I first saw your Control Bootcamp series and I simply can't seem to get enough of the knowledge that you have shared. It is really interesting the way you present these subjects. I will binge this one tonight :)
Sweet, I've been planning to get a small cheap car or a junkyard powertrain, and try my hand at fuel, ignition, and geometry controls, with emissions, efficiency, and responsiveness goals. There seems to be a lot of untapped potential there, given that engine ECUs are still largely based on low resolution maps and map overlays rather than explicit nonlinear controls and flexible memory. It may still be worthwhile to visualize data driven controls this way, but it seems like a limited way to author them. I hope there are some gems in here to get me in the right mindset for this. Thank you for sharing.
Hi, thank u for this overview. An observation from practical industrial experience: so far we see ML/deep learning has limited use cases, due to the need for clean, good data and difficulty to deal with dynamics. Most industrial data aren't images or tweets. ML and non-linear optimization is already part of some optimization systems, but this is modeling and statistics. For control we still will keep it simple from cost, safety and reliability perspective, as well as the brownfield difficulty that arises of implementing these in existing plants. I personally don't see the use cases of controlling neurology, mixing chems or turbulence in the near term as likely to utilize these approaches, as they either have existing, safe control solutions (usually linearized haha) or are use cases unlikely to be controlled. I really am trying to keep my mind open about the ML but so far I saw very few real life industrial use cases for control. I hope I'm proven otherwise. The last part on direct ML control sounds awesome and could be the real use case - but would require such in depth understanding of the system, which goes back to the data problem. Thanks again
Hello! I’m scratch building and programming an autonomous quadcopter to learn some of these methods in application. My goal eventually is to replicate the fast-pace actions seen in drone racing autonomously. Could I use some kind of MPC with data driven system identification for better model performance? My plan is to use a fast cheap control law like pid to enact angular setpoints, and a slower but smarter controller that determines what those angles should be. Do you have any recommendations? Thanks for the awesome content
I would like to build a mechatronic dragon or bat or bird that could learn how to set its wings against the wind and sail around using virtually no energy like real birds do.. literally all it has to do is deflect the wind downwards slightly and occasionally flap to stop itself being dragged along with the wind and that's SO MUCH more efficient than spinning rotors at thousands of rpm it could stay up for hours!
As an embedded systems engineer who studied control theory at graduate school and had machine learning oriented internships, I find myself today in my career at the crossroad between all 3 fields. I just wanted to say that your work has been a major help in igniting my passion for control theory back in school and now I am finding out that you're bringing machine learning into it. I look up to you and am just fascinated by the work you do making this knowledge available publicly. I always entertain the idea of meeting you someday and picking your brain on various topics and engineering problems.
Damn, now I really wished I'd applied to UWash! Prof. Brunton's teaching style really resonates with me. Thank you so much for putting these up!
Just want to shout out how good this series is. I came away with understanding and inspiration. Anyone here at the start, settle in and enjoy, Steve makes this a joy to learn.
Ppl like u r way forward for development and developers.... hats off!!
Found your series while looking for more information about data-driven modeling, machine learning and AI, as a software engineer interested in learning. The math initially made my head explode; it's been too many years since I reviewed algebra and worked through equations. I can feel my skull expanding as certain neural highways rebuild themselves while watching. Excellent lectures!
Thank you so much for your videos! I really enjoy the practical mix of controls, physics, and computing you present!
Glad you like them!
I found your video, because I searched for a solution to write a just in time interpreter for a just in time interpreter. Some kind of data driven subscript language on PHP. In the end it's a very high dynamic system. Your introduction brought me some vocabulary to think further. Thx a lot! 🤗
@Steve Brunton,I first saw your Control Bootcamp series and I simply can't seem to get enough of the knowledge that you have shared. It is really interesting the way you present these subjects. I will binge this one tonight :)
Sweet, I've been planning to get a small cheap car or a junkyard powertrain, and try my hand at fuel, ignition, and geometry controls, with emissions, efficiency, and responsiveness goals.
There seems to be a lot of untapped potential there, given that engine ECUs are still largely based on low resolution maps and map overlays rather than explicit nonlinear controls and flexible memory. It may still be worthwhile to visualize data driven controls this way, but it seems like a limited way to author them.
I hope there are some gems in here to get me in the right mindset for this. Thank you for sharing.
I am inspired. Thank you!
Im so binging this series today
I went to the University of Waterloo so when I saw databookuw I got really excited... looks like I went to the wrong UW
Steve, I have become a big fan of you!
I love how SBrunton says something like "Toddler that explores system and tries different control strategies" about just born kid :)
Hi, thank u for this overview. An observation from practical industrial experience: so far we see ML/deep learning has limited use cases, due to the need for clean, good data and difficulty to deal with dynamics. Most industrial data aren't images or tweets.
ML and non-linear optimization is already part of some optimization systems, but this is modeling and statistics. For control we still will keep it simple from cost, safety and reliability perspective, as well as the brownfield difficulty that arises of implementing these in existing plants.
I personally don't see the use cases of controlling neurology, mixing chems or turbulence in the near term as likely to utilize these approaches, as they either have existing, safe control solutions (usually linearized haha) or are use cases unlikely to be controlled. I really am trying to keep my mind open about the ML but so far I saw very few real life industrial use cases for control. I hope I'm proven otherwise.
The last part on direct ML control sounds awesome and could be the real use case - but would require such in depth understanding of the system, which goes back to the data problem.
Thanks again
I wonder if there are any changes to your opinions about a year after?
Really enjoy the lecture and the teaching style. Any thoughts on making a lecture series on Autonomous systems or acrtually Kalman Filters.
As a student , what are the job prospects if you work on a few projects in data driven control
Hello! I’m scratch building and programming an autonomous quadcopter to learn some of these methods in application. My goal eventually is to replicate the fast-pace actions seen in drone racing autonomously. Could I use some kind of MPC with data driven system identification for better model performance?
My plan is to use a fast cheap control law like pid to enact angular setpoints, and a slower but smarter controller that determines what those angles should be. Do you have any recommendations?
Thanks for the awesome content
Check this out : ruclips.net/video/FHvDghUUQtc/видео.html
How can I get ideas to solve problems with this approach?
Are you an expert at writing backwards or is this some sort of a video editing trick?
I'm glad someone asked!
Videos are flipped, he is actually left handed …
is there any blog/website for such topic on control systems ? can I have the links. thank you.
I am looking into compiling some references for this material.
good videos
Hi . have you a videos course based on adaptative neuro control or a related courses ? thank you .
I would like to build a mechatronic dragon or bat or bird that could learn how to set its wings against the wind and sail around using virtually no energy like real birds do.. literally all it has to do is deflect the wind downwards slightly and occasionally flap to stop itself being dragged along with the wind and that's SO MUCH more efficient than spinning rotors at thousands of rpm it could stay up for hours!
I really thought this was filmed post-COVID because of the pandemic example
Now when you hear "epidemic weaves" it hits harder ;/