--- Log opened Mon Jan 20 00:00:53 2025
01:58 < hprmbridge> nmz787> Fenn do you know any tools that i could whip up procedurally generated forest, and then build or import experimental robot designs, and then test their efficacy in said world?
02:21 < fenn> that sounds insanely complicated
02:23 < fenn> i mean these kinds of simulations exist, but getting them to run at all is a multi-day task
02:25 < fenn> e.g. https://gazebosim.org/showcase
02:32 < fenn> it would be fairly easy to scatter foresty objects around a terrain heightmap with a particle system in blender
02:32 < fenn> getting those to behave correctly in a physics sim is another matter
02:32 < fenn> like a fern is not infinitely rigid
02:36 < fenn> i wouldn't know where to begin with converting cad mechanisms to simulations
03:36 -!- gl00ten [~gl00ten@bl5-237-194.dsl.telepac.pt] has joined #hplusroadmap
03:43 -!- gl00ten [~gl00ten@bl5-237-194.dsl.telepac.pt] has quit [Ping timeout: 252 seconds]
05:23 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
05:51 < darsie> https://www.youtube.com/watch?v=-8SplPX2xVQ
05:51 < darsie>  The digital afterlife isn’t quite what you expected | Chris & Jack 
05:52 < darsie> scifi
06:47 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has joined #hplusroadmap
07:10 -!- gl00ten [~gl00ten@bl5-237-194.dsl.telepac.pt] has joined #hplusroadmap
07:12 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
07:12 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
07:44 < kanzure> nmz787: hod lipson probably has a robot simulation environment either of his own or he could recommend you one
09:33 -!- gl00ten [~gl00ten@bl5-237-194.dsl.telepac.pt] has quit [Ping timeout: 252 seconds]
10:26 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has quit [Quit: Under maintenance, I'll be back...]
10:49 < hprmbridge> nmz787> Hrmmm
10:51 < hprmbridge> nmz787> Even figuring out the center of gravity for some random tool on AliExpress or something would not be trivial if you didn't already have it to test
11:34 -!- Croran [~Croran@user/Croran] has joined #hplusroadmap
13:03 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has joined #hplusroadmap
14:21 < geneh2> genesis looks interesting for robot simulation https://genesis-embodied-ai.github.io/
14:24 < hprmbridge> kanzure> what's the one that nvidia says they have accelerated by 1000x or something
14:25 < geneh2> I don't think we can make very good models of a forest. Soil mechanics is pretty hard to model. You might be able to do some soil mechanics sim with genesis, but good luck making the sim fit real world dirt
14:25 < geneh2> Chrono engine's can supposedly model some pretty complicated stuff and has been used to investigate terramechanics of vehicles https://projectchrono.org/
14:26 < geneh2> Drake's supposed to be better at modeling contact than other physics engines: https://drake.mit.edu/
14:30 < geneh2> the truth is that rigid body physics engines aren't really good at modeling rigid bodies, especially for modeling robots picking stuff up. Drake's supposed to do better.
15:50 -!- gl00ten [~gl00ten@bl5-237-194.dsl.telepac.pt] has joined #hplusroadmap
16:47 < fenn> the contact force vector thing looks so obvious when you see it. https://miro.medium.com/v2/resize:fit:640/format:webp/1*mdPmUks_MHlszm2iMEqQbw.gif vs https://miro.medium.com/v2/resize:fit:640/format:webp/1*UdYHvqYcDx-4U7iBp5g2ow.gif
16:49 < fenn> makes you wonder how physics engines even work at all (maybe this is why they tend to suck)
17:02 < fenn> they pre-calculate a pressure field with Laplace's equation to approximate deformation near the contact point
17:03 < fenn> pressure is zero at the surface of the object and gradually increases toward the center
18:19 < geneh2> they work, they just don't work very well
19:37 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Ping timeout: 248 seconds]
19:53 -!- justanot1 [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
19:55 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 264 seconds]
20:26 < fenn> hold on, you're telling me a 7B model beats GPT-4o on ALL the benchmarks? https://old.reddit.com/r/LocalLLaMA/comments/1i5q6b9/deepseekr1_and_distilled_benchmarks_color_coded/
20:27 < fenn> all the listed benchmarks*
20:33 < superkuh> That testing makes claude look pretty bad. It is not bad.
20:34 < superkuh> Also: I want to believe.
21:22 < fenn> "It picks your code apart critically and unrelentlessly, every code smell, every bad practice, it points out and fixes. you can't hide a single thing from this motherf**ker"
22:15 < hprmbridge> kanzure> https://ssno.cc/posts/reversing-tac-1-4-2025/
23:16 < fenn> can we get an LLM to annotate all the variable names yet
23:21 < fenn> screenshotting and uploading users' windows without permisison is naughty
--- Log closed Tue Jan 21 00:00:54 2025