Nathan Yan

I built a drone VLA that navigates from natural language commands. Using Google’s Paligemma as an input head, and Pegasus Simulator's built-in drone simulation features like LiDAR, I was able to train a drone VLA to navigate to random objects in Isaac sim successfully.

@_reesechong goat

I added KV caching and INT8 KV quantization to our transformer inference, improving throughput by 35x. All of this was done from scratch in Rust + CUDA, on top of a homemade ML framework. On a 4-token prompt with 252 generated tokens: - Original: 0.76 tok/s - KV cache fp32: 27.21 tok/s - KV cache int8 (quantized): 27.29 tok/s Try it out yourself here: mni-ml.github.io/demos/kv-cache/ In practice: - KV caching gave us about a 35x end-to-end speedup - INT8 KV cache kept roughly the same speed as fp32 but cut KV cache memory by 3.78x FP32 cache used 4.5 MB in this run while the INT8 cache used only 1.19 MB This simple change to inference created a huge impact on performance. To learn more about the KV cache and other optimizations like this, check out the blog at mni.ml!

@PetarIsakovic06 @HackPrinceton yessss

@OfficialNathanY @HackPrinceton LOCKED IN!!!

Won 1k from @HackPrinceton. Back to VLAmaxxing!

@selimaz_ @HackPrinceton yessir

@OfficialNathanY @HackPrinceton hell yeah 🔥

@PetarIsakovic06 What a project!!

Built “ReciMe for clothes” send a reel → get every outfit + exact items instantly

hosting a @vercel @v0 build night this friday 4 hour build night use v0 platform to create something agentic free credits + merch happening at stan hq, downtown toronto reply or dm if you want in

@indefeasible_ @HackPrinceton 🙏🙏

@OfficialNathanY @HackPrinceton

@advay_c @HackPrinceton thank u goattt

@OfficialNathanY @HackPrinceton holy congrats dawg

@jamescowjam @HackPrinceton Yessirrr

@OfficialNathanY @HackPrinceton Lets fuking goooooo

@shayaan_azeem @HackPrinceton This is tuff

im at princeton judging @HackPrinceton today!! hmu if you're here, would love to say hi

1/ Introducing CLAW 🦀 — a pipeline for scalable generation of language-annotated whole-body motion data for the Unitree G1 humanoid. Joint work with @ThomasYuxinChen at MSC Lab, UCberkeley Code: github.com/JianuoCao/CLAW Tech Report: arxiv.org/pdf/2604.11251

Vision language action models finally get to look around on their own. Through the paper SaPaVe, researchers allowed robots to actively move their head-mounted camera to find objects completely out of view. (Day 2 of highlighting interesting CVPR 2026 papers about VLAs)

@rickyramosx trained a navigation MLP head in Isaac Lab and used that as the base model. Then attached Paligemma as for vision and language capabilities (basically took inspiration from Pi0) Need to check the amount of trajectories I trained it on lol dont have it top of head rn

@OfficialNathanY Super cool stuff, what was the base model and how many trajectories did you train on

I built a drone VLA that navigates from natural language commands. Using Google’s Paligemma as an input head, and Pegasus Simulator's built-in drone simulation features like LiDAR, I was able to train a drone VLA to navigate to random objects in Isaac sim successfully.

@IlirAliu_ Definitely a cool paper!

Robots don’t feel, they just follow positions. That’s why they fail in the real world. This paper shows what changes: It lets robots feel force while acting. Not just where to move, but how hard to push. That’s the difference between demo and real work. Thanks for sharing, @OfficialNathanY ! 📍Paper: arxiv.org/pdf/2603.15169 Project: sites.google.com/view/force-vla… Would love to hear what my friend @Klajd_Lika thinks about this. —— Weekly robotics and AI insights. Subscribe free: 22astronauts.com

@KrishivThakuria when mythos drops Ima see u coding in mountainpeaks bro

@OfficialNathanY Highkey true

Building on a flight is truly a peak experience

@ayushrgarg yesy

is ts tuff

paper: arxiv.org/pdf/2603.15169

Vision language action models finally get to feel what they're doing. Through the paper ForceVLA2, researchers allowed robots to grab and manipulate objects via force. (Day 1 of highlighting interesting CVPR 2026 papers about VLAs)

@michael_mzl @sentra_app LFGG

I’ll be joining @sentra_app in San Francisco as a Software Engineering Intern!!! If you’ll be in the Bay Area this summer, let’s connect. (Hmu, I need to find a roommate 😭)