Andrea Cognolato

Robotics, probabilistic ML, data systems

(opinions are my own and not the views of my employer)

[Andrea Cognolato @bayesbrah]

part 2 of my quadcopter series is out: going from 2D to 3D.

we derive step-by-step the Euler-Newton equation of motions, quaternion derivatives, and a state-space model ready to simulate

more in thread 🧵

[Andrea Cognolato @bayesbrah]

sneak peek of next blog post, finally multiple laps!

the quadcopter has 36 N / (0.78 kg * 9.81) = 4.7 thrust to weight ratio, higher than but comparable to Kaufmann et al's 4.1

life's easy since as we have perfect state estimation, but we'll switch to vision soon

[Andrea Cognolato @bayesbrah]

viz powered by the amazing @rerundotio

[Andrea Cognolato @bayesbrah]

part 2 of my quadcopter series is out: going from 2D to 3D.

we derive step-by-step the Euler-Newton equation of motions, quaternion derivatives, and a state-space model ready to simulate

more in thread 🧵

[Andrea Cognolato @bayesbrah]

HN discussion https://t.co/nwsRDYq6bI

in the next post, I'll make an RL environment using this simulator (or an optimized Rust version 🤫) and use PPO to train a racing policy

sneak peek:

[Andrea Cognolato @bayesbrah]

HN discussion: https://t.co/KQ7f5FzLVW

[Andrea Cognolato @bayesbrah]

After a long break, I'm back to blogging. Working on RL for (aerial) robotics.

First post: simulating a 2D quadcopter from scratch (equations of motion, state-space form, Python). Building the foundation before adding control. 🧵

[Andrea Cognolato @bayesbrah]

Blog post: https://t.co/bIkMaSiarF

More posts on control and RL coming soon.

[Andrea Cognolato @bayesbrah]

@AlicentShy91926 @SarikaS87472352 oh wow a scam account

[Andrea Cognolato @bayesbrah]

@TK_HelpDesk bro ghosted me

[Andrea Cognolato @bayesbrah]

@workwork42 reported

[Andrea Cognolato @bayesbrah]

@workwork42

[Andrea Cognolato @bayesbrah]

@0xmisaka @savetonotion #thread

[Keno Fischer @KenoFischer]

Congratulations to @nvidia for finally open sourcing their kernel drivers. Long time coming, but absolutely the right move. Looking forward to sending some PRs :).

[Logan Kilpatrick @OfficialLoganK]

Spotted the JuliaHub logo in Season 4 of #DriveToSurvive on @netflix 🏎. Skip to 8 minutes and 20 seconds left in episode 6!

Hopefully @williamsf1 ends up as fast as #JuliaLang! CC @JuliaComputing @Viral_B_Shah

[Andrea Cognolato @bayesbrah]

New post post on "predictive sampling and graph traversals". The natural continuation of the last post on probabilistic programming languages.
https://t.co/NDSlWeZBlR
#python #probabilisticprogramming #MachineLearning

[Andrea Cognolato @bayesbrah]

@jaakkolehtinen They have a "Download your data" tool in the account's privacy settings which can do it.
https://t.co/6tHw1LeFMM

[Andrea Cognolato @bayesbrah]

@ChadScherrer 1% of the range but not 1% of the values, right? Floats are not equally spaced so they are more dense around 0 I think

[Andrea Cognolato @bayesbrah]

@dustinvtran @mcspeedrunning Apparently a rebuttal has been written https://t.co/kd8rcpRMqb

It criticizes the modeling and uses a Monte Carlo simulation for a better Bartering probability estimation