Sanjana Srivastava

Large robot datasets are crucial for training 🤖foundation models. Yet, we lack systematic understanding of what data matters. Introducing MimicLabs ✅System to generate large synthetic robot 🦾 datasets ✅Data-composition study 🗄️ on how to collect and use large datasets 🧵1/

Job seekers are using LLMs to boost their resumes. Are companies interviewing the best candidates ... or just the candidates using the best LLM? 🧐 Our new ICML paper presents a fair and accurate hiring algorithm under stochastic manipulations: 📄 arxiv.org/abs/2502.13221 🧵 1/5

Proud to see the env I contributed to Maniskill being used by researchers in the community😆. Check the interesting work on utilizing unconstrained and dynamic preferences for reward generation!

ROSETTA, best paper at CRLH @ RSS. Generating rewards based on preferences with improved performance and alignment compared to past work like Eureka and Text2Reward

The @RoboticsSciSys conference concluded and I’m amazed by the growing number of projects using the ManiSkill framework! Glad to see how we enable others’ research in robotics, some have been nominated/won best paper awards! I’ve highlighted a few of them in the thread below 🧵

Foundation models can serve as "translators" btw humans and robots -- by translating unconstrained language instructions or feedback into reward functions for robot policy learning. Check out our recent work ROSETTA@RSS workshop!

From my understanding, ROSSETA is basically RL from human+AI feedback. Aligning robot with human preference will be a crucial piece of everyday robot. Great work!

Come chat with @mengdixu_ at HitLRL this afternoon!! I'll be giving a lightning talk at 10:15am as well :)

I’ve always been thinking about how to make robots naturally co-exist with humans. The first step is enabling them to understand our unconstrained, dynamic preferences and follow them. 🤝 We proposed ROSETTA, which translates free-form language instructions into reward functions…

How can we move beyond static-arm lab setups and learn robot policies in our messy homes? We introduce HoMeR, an imitation learning agent for in-the-wild mobile manipulation. 🧵1/8

(1/11) Evolutionary biology offers powerful lens into Transformers learning dynamics! Two learning modes in Transformers (in-weights & in-context) mirror adaptive strategies in evolution. Crucially, environmental predictability shapes both systems similarly.

BASED ✌️ turns 1! One year since its launch at NeurIPS 2023 — and it's helped shape the new wave of efficient LMs. ⚡️ Fastest linear attention kernels 🧠 405B models trained on 16 GPUs 💥 Inspired Mamba-v2, RWKVs, MiniMax Checkout our retrospective below!

BEHAVIOR is coming to life!! Check out this work led by @YunfanJiang that brings complex, meaningful tasks to the real world. I'm especially excited about the intuitive framework for scaling teleop data.

Started blogging again after 5 years. My first take: we should price LLMs in $ per bit, not in $ per token.