Chandradithya J

MS CS @BrownUniversity, RL, Robot Learning 🦾, prev: undergrad @iit_tirupati, intern @rbc_dsai_iitm

[Chandradithya J @enigmatic_chad]

@wesleymaa @hazel_heejeong Interesting question - I feel like you'd run into the obvious information vs mapping accuracy tradeoff so maybe you'd have to tune them together?

[Chandradithya J @enigmatic_chad]

@wesleymaa @hazel_heejeong Thanks, Wesley! We just followed previous works that choose 32 and also did a not-so-rigorous analysis where we scale down the #(trajectories) from 128 to 32 and compare performances.

[Hazel Nam @ ICML @hazel_heejeong]

Can a Latent Action Model really know what the “action” is?

In Super Mario, Mario 👨‍🦰 moves but so do 🍄, ☁️,🌳, and the view🎥.

TL;DR: under agent ambiguity, don’t force LAMs to find the true action directly. We factorize what changed.

Paper: https://t.co/nnEk4q222M

[Lakshita Dodeja @ RSS @lakshitadodeja]

Can BC policies be quickly improved through real world experience?

Our new #RSS2026 paper proposes Q2RL, a method that bridges BC and RL for on-robot learning.

Q2RL improves BC policies by up to 3.75x with just 1-2 hours of online interaction!

So when life gives you BC, make Q-functions! 🍋

Details in thread 🧵

[Chandradithya J @enigmatic_chad]

This work is done in collaboration with Deep Ganguly (first author), Pratham Chintamani, and Adithya Ananth.

Read the paper here: https://t.co/PePuTg0JwH

We'll be at the #ALA workshop at #AAMAS2026! Feel free to stop by and chat about MARL, Safe RL or long-horizon RL! (5/5)

[Chandradithya J @enigmatic_chad]

Why do civilizations collapse? We tackle this in multi-agent RL! (ALA @ AAMAS 2026) 🤖

When AI agents cooperate, exploration looks like betrayal. We introduce RATTL: a zero-communication trust mechanism, and define the new Price of Paranoia. #AAMAS2026 #ALA (1/5)

[Chandradithya J @enigmatic_chad]

Alongside RATTL, we also introduce the Price of Paranoia - the structural dual of the Price of Anarchy - to precisely characterize achievable cooperative welfare under partner noise. (4/5)

[Calvin Luo @calvinyluo]

How can visual planning agents 𝙨𝙚𝙡𝙛-𝙞𝙢𝙥𝙧𝙤𝙫𝙚 from their own collected experience?

We present 𝗦𝗜𝗟𝗩𝗥🩶, a framework that combines offline data with online experience for concurrent zero-shot generalization and sample-efficient self-improvement capabilities!#ICLR2026

[Chandradithya J @enigmatic_chad]

どんな #ポケモン会えるかな ？

#ポケモン30周年 #PR @poke_times

[Dan Haramati @DanHrmti]

Code, datasets and checkpoints are now available!

https://t.co/oBHlnRtk1Y

#WorldModels #Diffusers #OfflineGCRL #HierarchicalRL #EntityCentricRL

[Dan Haramati @DanHrmti]

Learning accurate World Models for long horizon planning is hard.

So what minimal aspect of world dynamics must a model capture to achieve complex goals?

We find a simple and effective solution in our #ICLR2026 paper, which we will present as an Oral at @worldmodel_26.

(1/n)