Excited to be at #ICML2026 this week presenting joint work with Chongyi Zheng and @ben-eysenbach.bsky.social on "Consistent Zero-Shot Imitation with Contrastive Goal Inference" (1/4)

🧠🔭Today's AI models synthesize knowledge acquired from the internet/books/etc. Ultimately, that knowledge usually derives from real experiments. We know (say) the moon's mass because a human did a science experiment. How well do AI models fare at generating knowledge? 🤔

Raj Ghugare@raj-ghugare.bsky.social · 4mo ago

I spent some time evaluating the best AI models on interactive block-building tasks. I am surprised by 1) the fragility of these systems when trying to generate creative ideas or update hypotheses, and 2) the vast, but often unnecessary, knowledge and compute they are willing to throw.

🤖Excited to share SLAP, @yijieisabelliu.bsky.social 's new algorithm using RL to provide better skills for planning! Check out the website for code, videos, and pre-trained models: github.com/isabelliu0/S...

GitHub - isabelliu0/SLAP

Contribute to isabelliu0/SLAP development by creating an account on GitHub.

github.com

@yijieisabelliu.bsky.social · 9mo ago

Robots can plan, but rarely improvise. How do we move beyond pick-and-place to multi-object, improvisational manipulation without giving up completeness guarantees? (1/8)

Kids spend years playing with blocks, building spatial+arithmetic skills. Today, AI models just read. While AI research often conflates reasoning with language models, block-building lets us study how embodied reasoning might emerge from exploration and trial-and-error learning!

Raj Ghugare@raj-ghugare.bsky.social · 10mo ago

Can AI models build a world which today's generative models can only dream of? Presenting BuilderBench (website : t.co/H7wToslhXG). Details below 🧵⬇️

🚨 Excited to announce our #NeurIPS2025 Workshop: Data on the Brain & Mind 📣 Call for: Findings (4- or 8-page) + Tutorials tracks 🎙️ Speakers include @dyamins.bsky.social @lauragwilliams.bsky.social @cpehlevan.bsky.social 🌐 Learn more: data-brain-mind.github.io

Check out @raj-ghugare.bsky.social's new paper on the surprising effectiveness of normalizing flows (NF) in RL 🚀 This project changed my mind in 2 ways: 1/ Diffusion policies, flow-models, and EBMs have become ubiquitous in RL. Turns out NFs can perform as well -- no ODEs/SDEs required!

Raj Ghugare@raj-ghugare.bsky.social · last yr.

Normalizing Flows (NFs) check all boxes for RL: exact likelihoods (imitation learning), efficient sampling (real-time control), and variational inference (Q-learning)! Yet they are overlooked over more expensive and less flexible contemporaries like diffusion models. Are NFs fundamentally limited?

tldr: increase the depth of your RL networks by several orders of magnitude. Our new paper shows that very very deep networks are surprisingly useful for RL, if you use resnets, layer norm, and self-supervised RL! Paper, code, videos: wang-kevin3290.github.io/scaling-crl/

@kevin-wang3290.bsky.social · last yr.

1/ While most RL methods use shallow MLPs (~2–5 layers), we show that scaling up to 1000-layers for contrastive RL (CRL) can significantly boost performance, ranging from doubling performance to 50x on a diverse suite of robotic tasks. Webpage+Paper+Code: wang-kevin3290.github.io/scaling-crl/

Excited to share new work led by @vivekmyers.bsky.social and @crji.bsky.social that proves you can learn to reach distant goals by solely training on nearby goals. The key idea is a new form of invariance. This invariance implies generalization w.r.t. the horizon.

Vivek Myers@vivekmyers.bsky.social · last yr.

Reinforcement learning agents should be able to improve upon behaviors seen during training. In practice, RL agents often struggle to generalize to new long-horizon behaviors. Our new paper studies *horizon generalization*, the degree to which RL algorithms generalize to reaching distant goals. 1/