#6.832blogging | Explore Tumblr Posts and Blogs

gender-trash · 5 months

Text

the second funniest thing that happened to me when i took underactuated robotics was when my dad (who was watching the lecture recordings on youtube as they came out) sent me a low-res youtube screenshot of the side of my face as i was sitting down in my chair. stalker behavior??

(the funniest thing that happened to me was the time he showed a sim video of my incredibly janky implementation of marc raibert's one-legged hopping robot controller falling over TO MARC RAIBERT HIMSELF. dad!! i sent you my 3am controller tuning fail content in confidence!!!! (i'm not really mad about this anymore, but don't tell him i said that bc i want to keep razzing him about it))

(EDIT: ...this is probably funnier if u are aware that marc raibert is the ceo of boston FUCKING dynamics.)

#the trashcan speaks #6.832blogging #professional robot torturer

23 notes · View notes

gender-trash · 5 months

Note

I had to go to a department seminar today for a requirement and the talk was 'Modeling, Estimation, and Control of Quadrupedal and Humanoid Robot Locomotion in Non-Inertial Environments' and it was interesting but I kept thinking 'darn, I bet gender-trash would love this' It was essentially, hey can we get a robot to walk in a straight line while on a rocking ship?' and the answer was 'lol. sorta kinda' anyhow I hope you have a nice evening :)

:0 this is really cool!! i looked up the person giving the seminar (yan gu) and found this video linked from one of the papers she coauthored:

youtube

(n.b. that i just skimmed this paper and am definitely not bothering to work through all the math, but) the key assumptions this makes are that 1) the robot can perceive or otherwise know the movement of the surface relative to itself (in the lab they use fiducials stuck to the treadmill, and hint vaguely in the paper that integrating sensor data from the ship or whatever would be plausible in a real-world deployment, which -- PERSONALLY i am much less confident about, as someone who has been socially adjacent to industry work on robot integration with elevator controllers, but whatever, it's research, i'm willing to cut them slack on that) and 2) the walking surface is planar (for the biped the surface position is determined from forward kinematics assuming the robot's feet always make full contact with the surface when it does a steppy; i read the quadruped paper much less thoroughly but given how careful they were to select a gait that always has three feet in contact with the ground i'm assuming the same condition holds).

DEFINITELY a big improvement over "let's assume the floor does not move :)" control, and i don't mean to be critical here, i think it's perfectly fair for a controls paper to leave the software integration/perception challenges up to Future Work to figure out! this is just how i personally analyze robotics research -- there's always a bunch of assumptions involved to make a robot problem tractable, some of them more realistic than others, so the most important thing i want to understand about a new research thing is what assumptions they're making. (for example, a lot of navigation research assumes no sane person would design or construct a building like the stata center...)

(also, really makes you appreciate how average humans can walk on wildly pitching ships with zero perception, just pure IMU, surface contact sensing, and proprioception/kinematics. take a moment to be grateful for your cerebellum :p)

anyway -- thanks for the pointer, i always love seeing awesome new controls-y stuff! anon, i assume you know most of this stuff already, but for followers interested in learning more about controls for leggy bois, i always recommend russ tedrake's underactuated robotics class -- he has his very well-written lecture notes and several years worth of lecture videos all available online. thanks prof tedrake i love you <3

#the trashcan speaks #6.832blogging #they fucking renumbered everything but you know what i mean #professional robot torturer

9 notes · View notes