artkuo.bsky.social
@artkuo.bsky.social
I'm sure you're a great speaker but I generally glaze over during "lemme tell you a bit about my other work." I just wanna hear a story. I suppose it is armor against criticism like "yeah but can candidate do anything else?" but I just dislike "job" add-ons.
March 21, 2025 at 9:41 PM
I'm sure you're a great speaker but I generally glaze over during "lemme tell you a bit about my other work." I just wanna hear a story. I suppose it is armor against criticism like "yeah but can candidate do anything else?" but I just dislike "job" add-ons.
They never used my idea :( 8/end
March 20, 2025 at 7:35 PM
They never used my idea :( 8/end
Yrs ago during a conference break, I saw BD crew sitting around talking ideas for new Atlas vids. I had a great idea: Spoof the spoof vids of person kicking 2 humans pretending to be BigDog. Have Atlas kick a BigDog! (They were & prob still are separate teams). Or kick 2 Atlases posing as BigDog. 7/
March 20, 2025 at 7:35 PM
Yrs ago during a conference break, I saw BD crew sitting around talking ideas for new Atlas vids. I had a great idea: Spoof the spoof vids of person kicking 2 humans pretending to be BigDog. Have Atlas kick a BigDog! (They were & prob still are separate teams). Or kick 2 Atlases posing as BigDog. 7/
In past BD hasn't seemed to work fast, but now they reveal they can. Also cool about both outfits is how creative they are with videos. 6/
March 20, 2025 at 7:35 PM
In past BD hasn't seemed to work fast, but now they reveal they can. Also cool about both outfits is how creative they are with videos. 6/
In any case, a great response to Unitree, also doing amazing stuff. I don't think they talk much, but they are in a sense kindred spirits. Unitree has also been no-nonsense, simple concepts made viable w/ great tools. Their human reference stuff also mind-blowing. 5/
March 20, 2025 at 7:35 PM
In any case, a great response to Unitree, also doing amazing stuff. I don't think they talk much, but they are in a sense kindred spirits. Unitree has also been no-nonsense, simple concepts made viable w/ great tools. Their human reference stuff also mind-blowing. 5/
I suspect the RL is used to manage limbs around an explicit (not latent) COM. The human reference seems to be used well, looks amazingly fluid. Also multi-limb contact is incredible, doesn't reveal the step-step tapping of Raibert controller sometimes still evident in Spot. 4/
March 20, 2025 at 7:35 PM
I suspect the RL is used to manage limbs around an explicit (not latent) COM. The human reference seems to be used well, looks amazingly fluid. Also multi-limb contact is incredible, doesn't reveal the step-step tapping of Raibert controller sometimes still evident in Spot. 4/
Superficially may seem like major paradigm shift, like old-school Disney animation going rotoscope. BD's always been good at scripting amazing feats like parkour, but it's never been hand-coded one-off. Always been principled, COM first. As opposed to just a buncha actuators. 3/
March 20, 2025 at 7:35 PM
Superficially may seem like major paradigm shift, like old-school Disney animation going rotoscope. BD's always been good at scripting amazing feats like parkour, but it's never been hand-coded one-off. Always been principled, COM first. As opposed to just a buncha actuators. 3/
Kinda ironic cuz Robotics & AI Institute! Anyway they've embraced it now & looks great. But I bet the RL is not end-to-end & there's still a model & concept of center of mass in there. Maybe more like end-to-COM-to-end. Otherwise Marc Raibert would cease to exist. 2/
March 20, 2025 at 7:35 PM
Kinda ironic cuz Robotics & AI Institute! Anyway they've embraced it now & looks great. But I bet the RL is not end-to-end & there's still a model & concept of center of mass in there. Maybe more like end-to-COM-to-end. Otherwise Marc Raibert would cease to exist. 2/
I now believe Robo-Tuna was a self-hoax, maybe they truly believed it, didn't understand their measurements. Dug themselves a hole defending their "theory" though. Nevertheless made its mark on underwater robots, even if initial claims nonsense.
Science is done by humans, both fun & agony. 14/end
Science is done by humans, both fun & agony. 14/end
March 19, 2025 at 8:23 PM
I now believe Robo-Tuna was a self-hoax, maybe they truly believed it, didn't understand their measurements. Dug themselves a hole defending their "theory" though. Nevertheless made its mark on underwater robots, even if initial claims nonsense.
Science is done by humans, both fun & agony. 14/end
Science is done by humans, both fun & agony. 14/end
BTW this is not only a fun topic but also emotional. Yrs ago saw lecture about sailing, speaker could barely spit out how much he hated Bernoulli. (It's used wrong for explaining lift, but principle isn't wrong!) My fluids colleague got really worked up how much he hated MIT Robo-Tuna! 13/
March 19, 2025 at 8:23 PM
BTW this is not only a fun topic but also emotional. Yrs ago saw lecture about sailing, speaker could barely spit out how much he hated Bernoulli. (It's used wrong for explaining lift, but principle isn't wrong!) My fluids colleague got really worked up how much he hated MIT Robo-Tuna! 13/
Maybe in another hour he could describe a qualitative but operational theory that one could really work with, meaning know what to run CFD/experiments on.
Also wish Veritassium would revisit topic & do fuller & more correct video on it. 12/
Also wish Veritassium would revisit topic & do fuller & more correct video on it. 12/
March 19, 2025 at 8:23 PM
Maybe in another hour he could describe a qualitative but operational theory that one could really work with, meaning know what to run CFD/experiments on.
Also wish Veritassium would revisit topic & do fuller & more correct video on it. 12/
Also wish Veritassium would revisit topic & do fuller & more correct video on it. 12/
BTW I took fluids as grad stud (my UG was EE), only b/c I thought embarrassing to be PhD ME w/o knowing the first thing. Except I still don't know first thing, still embarrassed!
Babinsky's lecture ends w/ brief figures showing pressure far above & below wing. Now we're cooking! 11/
Babinsky's lecture ends w/ brief figures showing pressure far above & below wing. Now we're cooking! 11/
March 19, 2025 at 8:23 PM
BTW I took fluids as grad stud (my UG was EE), only b/c I thought embarrassing to be PhD ME w/o knowing the first thing. Except I still don't know first thing, still embarrassed!
Babinsky's lecture ends w/ brief figures showing pressure far above & below wing. Now we're cooking! 11/
Babinsky's lecture ends w/ brief figures showing pressure far above & below wing. Now we're cooking! 11/
Nonsensical how that's tucked into fluids and not part of all engineering & science. Another problem is that UGs who decide on fluids for grad school have no idea what it'll be. Surprise, your life is upside down triangles (div grad curl)! 10/
March 19, 2025 at 8:23 PM
Nonsensical how that's tucked into fluids and not part of all engineering & science. Another problem is that UGs who decide on fluids for grad school have no idea what it'll be. Surprise, your life is upside down triangles (div grad curl)! 10/
Also looking at old UG text, interesting they almost entirely avoid upside-down triangles for "accessibility". N-S barely mentioned. But the book is a slog, lots of scary numbers in eqns. (Common UG textbook malady: too many numbers.) Only worthwhile thing for me was dimensional analysis. 9/
March 19, 2025 at 8:23 PM
Also looking at old UG text, interesting they almost entirely avoid upside-down triangles for "accessibility". N-S barely mentioned. But the book is a slog, lots of scary numbers in eqns. (Common UG textbook malady: too many numbers.) Only worthwhile thing for me was dimensional analysis. 9/
The good news is there's plenty of room for Airfoil Whisperers, like Alan Adler of Aerobie fame. Some people just "get" boomerangs & can design & make them in infinite variations. But bad news is how little of that art has translated to engineering, still gotta put cyclist in wind tunnel! 8/
March 19, 2025 at 8:23 PM
The good news is there's plenty of room for Airfoil Whisperers, like Alan Adler of Aerobie fame. Some people just "get" boomerangs & can design & make them in infinite variations. But bad news is how little of that art has translated to engineering, still gotta put cyclist in wind tunnel! 8/
I believe fluidists still need a compact, operational theory that helps F1 engineers decide what to run CFD on. The popular L&D explanations are unfortunate b/c they sound like science & propagate some myths. Still, why is the book on winglets still being written—good theory worth many $$? 7/
March 19, 2025 at 8:23 PM
I believe fluidists still need a compact, operational theory that helps F1 engineers decide what to run CFD on. The popular L&D explanations are unfortunate b/c they sound like science & propagate some myths. Still, why is the book on winglets still being written—good theory worth many $$? 7/
Babinsky does great job but not enough to help F1 engineer. Also note that winglets are 50 yrs old and still refined/debated, do they still lack a complete theory? My interp is that N-S knows all, but too complicated to optimize a design; can only tweak a limited parameter range. 6/
March 19, 2025 at 8:23 PM
Babinsky does great job but not enough to help F1 engineer. Also note that winglets are 50 yrs old and still refined/debated, do they still lack a complete theory? My interp is that N-S knows all, but too complicated to optimize a design; can only tweak a limited parameter range. 6/
Nevertheless I think/hope someone who learns p.411-430 well could be useful to an F1 team? The presented theory is experimental (NACA) & principles, so shouldn't lead one too far astray. But I wonder if any theory is truly correct & operational, or is there only N-S? 5/
March 19, 2025 at 8:23 PM
Nevertheless I think/hope someone who learns p.411-430 well could be useful to an F1 team? The presented theory is experimental (NACA) & principles, so shouldn't lead one too far astray. But I wonder if any theory is truly correct & operational, or is there only N-S? 5/
I remember some great discussions w/ my fluids colleagues. I once asked them about fish, and they suffered a little talking about "what you (uncouth savages) call drag." That's cuz modern fluidists live in Navier-Stokes & often consider drag "popular science" & not a way to think. 4/
March 19, 2025 at 8:23 PM
I remember some great discussions w/ my fluids colleagues. I once asked them about fish, and they suffered a little talking about "what you (uncouth savages) call drag." That's cuz modern fluidists live in Navier-Stokes & often consider drag "popular science" & not a way to think. 4/
I took fluids back in the day and came out of it knowing nothing about lift & drag?! Looking back at textbook, L&D appear p.411-430 on external flows, incl airfoil theory. most of the other 700 pp on other stuff. So no wonder, L&D aren't really formal education, more armchair physics. 3/
March 19, 2025 at 8:23 PM
I took fluids back in the day and came out of it knowing nothing about lift & drag?! Looking back at textbook, L&D appear p.411-430 on external flows, incl airfoil theory. most of the other 700 pp on other stuff. So no wonder, L&D aren't really formal education, more armchair physics. 3/
Babinsky beautifully destroys the common explanations e.g. streamlines rush to meet each other. His is probably best of the qualitative theories, but part of the problem is that most are still non-operational descriptions: U can't use them to make useful predictions. 2/
March 19, 2025 at 8:23 PM
Babinsky beautifully destroys the common explanations e.g. streamlines rush to meet each other. His is probably best of the qualitative theories, but part of the problem is that most are still non-operational descriptions: U can't use them to make useful predictions. 2/
Can video-based autonomy drive as well as humans? Probably eventually w/ really good world models. But if current rate ~1.3 deaths/100M miles, then goal should be 0.13 or 0.013/100M. Likelier to get there w/ real depth sensors. LIDAR is $$, but so were early airbag sensors.
March 18, 2025 at 7:18 PM
Can video-based autonomy drive as well as humans? Probably eventually w/ really good world models. But if current rate ~1.3 deaths/100M miles, then goal should be 0.13 or 0.013/100M. Likelier to get there w/ real depth sensors. LIDAR is $$, but so were early airbag sensors.