KyleMorgenstein Profile Banner
Kyle🤖🚀🦭 Profile
Kyle🤖🚀🦭

@KyleMorgenstein

Followers
16K
Following
196K
Media
2K
Statuses
35K

Full of childlike wonder. Teaching robots manners. UT Austin PhD candidate. 🆕 RL Intern @ Apptronik. Past: Boston Dynamics AI Institute, NASA JPL, MIT ‘20.

he/him
Joined September 2018
Don't wanna be here? Send us removal request.
@KyleMorgenstein
Kyle🤖🚀🦭
1 year
when you argue with me about control theory this is who you’re arguing with
Tweet media one
5
5
111
@KyleMorgenstein
Kyle🤖🚀🦭
12 hours
Tweet media one
0
0
2
@KyleMorgenstein
Kyle🤖🚀🦭
21 hours
when I first read the paper I thought it could contend for a best paper award and this reviewer is arguing that it’s a strong reject because <redacted stupid reason>. just completely nuts.
1
0
15
@KyleMorgenstein
Kyle🤖🚀🦭
21 hours
on OpenReview are you allowed to comment back to other reviewers “that’s insane lmao” because one of them on a paper I’m reviewing is… literally insane.
4
0
35
@KyleMorgenstein
Kyle🤖🚀🦭
23 hours
RT @EugeneVinitsky: If you have the luxury to choose to do good I think you sort of have noblesse oblige to do so. If you can't, who else c….
0
1
0
@KyleMorgenstein
Kyle🤖🚀🦭
1 day
Very timely. Read this!.
@breadli428
Chenhao Li
1 day
🧠With the shift in humanoid control from pure RL to learning from demonstrations, we take a step back to unpack the landscape. 🔗 🚀Excited to share our blog post on Feature-based vs. GAN-based Learning from Demonstrations—when to use which, and why it.
1
1
6
@KyleMorgenstein
Kyle🤖🚀🦭
2 days
who is brave enough to turn @WolfeyGlick into a mechanical turk….
1
0
2
@KyleMorgenstein
Kyle🤖🚀🦭
2 days
it would be cool to have a compute-normalized ELO, to see the comparison in performance with different compute and parameter tiers, etc.
1
0
4
@KyleMorgenstein
Kyle🤖🚀🦭
2 days
maybe I can finally make my quagsire-gastrodon mono-water dreams come true… really excited to see what approaches end up being successful!.
@sethkarten
Seth Karten
2 days
🚀 Launch day! The NeurIPS 2025 PokéAgent Challenge is live. Two tracks:.① Showdown Battling – imperfect-info, turn-based strategy.② Pokemon Emerald Speedrunning – long horizon RPG planning.5 M labeled replays • starter kit • baselines. Bring your LLM, RL, or hybrid
Tweet media one
3
1
27
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
for robotics, tooling + infra >>> raw sim speed most of the time. (and with warp everyone is about to sim a lot faster too!).
0
0
8
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
inb4 “it’s slow as shit” ya we’re doing photorealistic rendering and full multi body dynamics at high enough fidelity to transfer to hardware. “speed” for us isn’t measured in sps alone, its iteration time for the human to close the feedback loop with testing on hardware.
2
0
4
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
Most people struggle because their URDF and gains aren’t as good as they think they are. But it’s not usually the RL that’s actually limiting them. It’s the infrastructure and system model that bite you in the butt.
1
0
6
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
You give me a new robot with a decent URDF and PD gains and a bunch of GPUs, I’ll get it working by the end of the day. I’m only half kidding.
2
0
9
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
I know this is literally Joseph-bait but please forgive me, it is true. Isaac is maybe not how you would design an RL methods test bed, but for deploying on real hardware fast, it’s really really good.
3
0
8
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
I know some software purists disagree but IsaacLab’s manager-based system is this. If you goal is to get RL working on hardware very fast, IsaacLab exposes all the right knobs without sacrificing the ability to hack it and add a huge amount of customization for your use case.
@jxmnop
jxmo
3 days
nice trend over the last year is that folks in AI have finally produced a few libraries with the right abstractions. finally our code can be both hackable and fast, not just one or the other. this never used to happen. vLLM, sglang, verl. this is the dawn of Good Software in AI.
5
0
34
@KyleMorgenstein
Kyle🤖🚀🦭
3 days
RT @observie: Pure gold from @KyleMorgenstein . Setting soft PID gains would allow an RL policy to achieve compliance, if needed, and when….
0
3
0
@KyleMorgenstein
Kyle🤖🚀🦭
4 days
it’s weird doing robotics in an aerospace department because I’ll be looking for quiet space in the building and accidentally walk into a mock satellite mission control or a half-built airplane.
5
0
79
@KyleMorgenstein
Kyle🤖🚀🦭
4 days
Not enough labs are taking advantage of how great a platform Spot is. I’m surprised so few use it, given its reliability and API.
0
0
4
@KyleMorgenstein
Kyle🤖🚀🦭
4 days
Super cool work! I have some similar methods coming out this fall using tactile estimation for various control tasks :).
@Hongyu_Lii
Hongyu Li
5 days
We interact with dogs through touch -- a simple pat can communicate trust or instruction. Shouldn't interacting with robot dogs be as intuitive? Most commercial robots lack tactile skins. We present UniTac: a method to sense touch using only existing joint sensors!.[1/5]
2
0
17
@KyleMorgenstein
Kyle🤖🚀🦭
6 days
RT @chris_j_paxton: @gietema Thanks! You can follow @TheHumanoidHub for robotics news . @ChongZitaZhang @KyleMorgenstein post lots of good….
0
6
0
@KyleMorgenstein
Kyle🤖🚀🦭
8 days
RT @EugeneVinitsky: Personally I would have trouble getting up in the morning if my job was "make sure the bot can be antisemitic" but that….
0
1
0