Sai ✈️ CoRL ‘26

co-founder stealth (@a16z @speedrun) | prev @tomadotcom | ex-pro gamer | ex-@uwaterloo

[Sai ✈️ CoRL ‘26 @saisurisetti_]

damn. even clav is on cc 💔

[Clavicular @clavicular]

I hate when a girl flexes me dming her on instagram, like bitch you're talking to claude code.

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@servo_boyd @corl_conf See u there!

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@Parvkpr Excited for this ;)

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@reactorworld @Ultraroboticsco Looking forward to it!

[Thariq @trq212]

this is basically the bitter lesson at work

the smartest models will need fewer and fewer tokens, they won't need to verify their work or check it for bugs, they will just be right

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@trq212 bittersweet

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@_varunnair no super strong competitors rn, but will probably see the rise of them as robotic fleets scale with all the deployments happening rn 😋

[Sai ✈️ CoRL ‘26 @saisurisetti_]

Can’t wait to benchmark models on their painting skills

[nil @dhvanil]

last night’s watercolor session with Claude! getting much better at trees now.

it also has a much bigger repertoire of brush strokes now, which enables it to compose drawings like this. and in addition, some cool new titration tricks.

Opus then used those brush movements to compose and paint “leeward”, an autumn tree leaning into the wind.

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@YijiaFanCSAI This is amazing dude

[Sai ✈️ CoRL ‘26 @saisurisetti_]

Yummy! Can’t wait for the upcoming wave of open source models specialized for CaP

[Yijia Fan @YijiaFanCSAI]

Every strong code-as-policy robot agent today runs on a closed frontier model. You can’t fine-tune it, you can’t put it on the robot, and it costs $8–22 per episode. Drop an open 4B model into the same harness and it solves nothing: 0.0% on LIBERO.

So can we post-train a small open model into a robot coding agent, the same way coding and tool-use agents are trained?

We’re excited to share CaPtain, an end-to-end pipeline for post-training Code-as-Policy robot agents.

The agent writes code against 22 low-level primitives (perceive, measure, move, check), all the way down to IK-level motion. There are no grasp() skills and no VLA underneath. We teach it with SFT, then sharpen it with on-policy RL using a critic that reads the simulator’s physics state, information the policy never sees.

The results:

Qwen3.5-4B on held-out LIBERO goes 0.0% → 55.4% (SFT) → 73.8% (RL). RL adds +17 to +21 points across LIBERO-PRO, LIBERO-Plus and LIBERO-MAX.
RL ran only on LIBERO, and the model carries over to other benchmarks with no further training: RoboTwin 2.0 Hard 22.6% (above X-VLA’s 20.9%), SimplerEnv WidowX 83.3%, CaP-Bench 75.0, plus RoboDojo.
The same API and the same model drive a real arm with no real-world data.

The craziest part: none of what the model learned is tied to an action space. It learned to look, measure, move and check, so the embodiment lives below the API.

Everything is open under MIT: weights, harness, SFT data, RL code and evals. It’s already at ~1.5K downloads on HF 🔥

We’re on HF Daily Papers today, so an upvote would mean a lot 🙏

📄 Paper: https://t.co/uomNje4xVB
💻 Code: https://t.co/61TTD3qCim
✍️ Blog: https://t.co/G74LlwzMa0

How it works, what broke, and robot videos below 🧵 (1/N)

[Sai ✈️ CoRL ‘26 @saisurisetti_]

There is an imposter…

[gautham ✈️ CoRL ‘26 @gautham_city_]

i can’t even tell who the billboard is for 😭

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@jerthesquare_ @dreamlabs_ai @gautham_city_ check this out lol

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@jerthesquare_ @dreamlabs_ai haha working on this exact thing rn, works a lot better than some other vlas as tools ;)

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@saikodakalla19 nice name

[Sai ✈️ CoRL ‘26 @saisurisetti_]

it lowk have human ahhh mannerisms

[Generalist @GeneralistAI]

How many pucks could a puck bot pack if a puck bot could pack pucks? 🏒

12 hockey pucks is a tight fit in this container – watch the model fix the arrangement when pucks get stuck or misplaced.

GEN-1.5 doesn't just puck around. Read more about it in our blog post and sign up for the waitlist in the comments below.

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@xdotli @dreamlabs_ai LOL that was quick, nice!

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@martinsit

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@Bobliuuu @AThoduvakkal Which is not good for a newly turned adult in a world where your opinions are echochambered on all socials

[Sai ✈️ CoRL ‘26 @saisurisetti_]

Bro is complaining about Waterloo being toxic while actively hating on a first year.

@AThoduvakkal’s point was that “cracked” upper years are generally helpful to younger students.

I agree, and I’d go even further. The only people I’ve ever seen being toxic are insecure losers. They’re either so scared of being outdone by another student that they feel the need to sabotage them, or they see people doing better and pick on younger, less experienced students to feel better about themselves. These people are always at most slightly above avg and get an ego about it.

@Bobliuuu, you’re talking about the latter. You should consider that the snakes you’re talking about might not be all that accomplished and are just hating because they’re insecure about themselves.

You should also reflect on your own behaviour. You’re calling out toxicity, so at least have the decency to be part of the change yourself.

[Jerry @Bobliuuu]

why are first years going on here to larp about how good waterloo culture is
comparing school to hackathon is so funny LMAO
"did yk cars are faster than macbooks"
you slowly realize that there's a snake in the grass but the snake does not reveal itself immediately
at least snakes at waterloo aren't that stupid

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@Bobliuuu @AThoduvakkal Making kids schizo about their peers is the best way to make them socially isolated

[Sai ✈️ CoRL ‘26 @saisurisetti_]

@Bobliuuu @AThoduvakkal I can’t read I’m dyslexic