Pinned Tweet
Do RL solutions share a common structure? We show that all solutions of Reinforcement Learning lie on a hyperplane. Our work, Proto Successor Measure, learns this abstraction of the MDP to do zero-shot RL for any reward function. (1/n)
Siddhant Agarwal retweeted
⏰ More time to submit! ⏰
We hope your drafts are getting into shape and we got good news! The #oopsiedata workshop @corl_conf 🤖 deadline is now October 5th. Polish up your work and send it our way! We are looking forward to all the excellent submissions!
🤖 Coming to @corl_conf 2026: The first workshop on imperfect robotics data! We’ll explore how to collect, curate & use failures 💢, OOD events❓, and unexpected human behavior 🤔 - data every lab produces but rarely documents, shares, or reuses. Check out workshop.oopsie-data.com
Siddhant Agarwal retweeted
Working on robot stuff but things didn’t work out for @ieee_ras_icra or @iclr_conf? Don’t fret, your results and data don’t have to go to waste.
At @corl_conf, we’re hosting a workshop on imperfect data. What can we do when things don’t go as planned?
🔗 workshop.oopsie-data.com/
Join us @corl_conf 2026 for the first workshop on imperfect robotics data to discuss how to collect and use failure demonstrations.
🤖 Coming to @corl_conf 2026: The first workshop on imperfect robotics data! We’ll explore how to collect, curate & use failures 💢, OOD events❓, and unexpected human behavior 🤔 - data every lab produces but rarely documents, shares, or reuses. Check out workshop.oopsie-data.com
Siddhant Agarwal retweeted
Two BIG updates for the RLBrew Workshop at #RLC2026! 📣
1️⃣ Dual submissions are welcome
2️⃣ We’ll be awarding a Best Paper RLBrew Award 🏆
You have 2 DAYS LEFT to submit — deadline: May 29!
Details: rlbrew-workshop.github.io/
Siddhant Agarwal retweeted
We are extending the submission deadline to May 29, 2026 (AoE). Submit your ideas soon!
Scalar, well-defined, easy-to-optimize rewards aren’t always available in real-world interaction data–yet that data is crucial for scaling general-purpose agents. Excited to announce the 3rd edition of RLBRew: Towards Scalable General-Purpose Agents at @RL_Conference 2026!
Can regularized latent dynamics prediction scale Behavior Foundation Models? Check out our work at ICLR 2026.
Introducing RLDP – a simple, scalable approach for building strong Behavioral Foundation Models. 🚀 #ICLR2026
✅ Robust objective: avoids brittle unsupervised RL objectives while staying simple and scalable.
🗂️ No data-coverage limitations: works across a wide variety of datasets.
🧩 One latent space for all tasks: a unified representation space.
🎯➡️🤖 Zero-shot control: specify a reward and directly obtain a policy/behavior – no additional training required.
I’ll be attending #NeurIPS2025 and presenting our work, “RLZero: Direct Policy Inference from Language Without In-Domain Supervision." Excited to chat about unsupervised RL, reasoning, and RL more broadly. I’m also exploring industry opportunities — feel free to reach out!
Siddhant Agarwal retweeted
Intelligent humanoids should have the ability to quickly adapt to new tasks by observing humans
Why is such adaptability important?
🌍 Real-world diversity is hard to fully capture in advance
🧠 Adaptability is central to natural intelligence
We present MimicDroid 👇
🌐 ut-austin-rpl.github.io/Mimi…
I’ll be at #ICML2025 presenting our paper, “Proto Successor Measure: Representing the Behavior Space of an RL Agent”. Excited to connect with others working on unsupervised RL and RL more broadly. Also am on the lookout for research collaborations and opportunities in industries.
Siddhant Agarwal retweeted
Most assistive robots live in labs.
We want to change that.
FEAST enables care recipients to personalize mealtime assistance in-the-wild, with minimal researcher intervention across diverse in-home scenarios.
🏆 Outstanding Paper & Systems Paper Finalist @RoboticsSciSys
🧵1/8
Siddhant Agarwal retweeted
Say ahoy to 𝚂𝙰𝙸𝙻𝙾𝚁⛵: a new paradigm of *learning to search* from demonstrations, enabling test-time reasoning about how to recover from mistakes w/o any additional human feedback! 𝚂𝙰𝙸𝙻𝙾𝚁 ⛵ out-performs Diffusion Policies trained via behavioral cloning on 5-10x data!
Siddhant Agarwal retweeted
We are extending the submission deadline to June 15, 2025 (AoE). Submit your ideas soon!
⚠️ Reminder! Submissions for @RL_Conference's RL beyond Reward Workshop are due May 30 (AoE)!
We are brewing an interesting program and seeking innovative research work in reward-free RL. All papers are welcome, from exploratory abstracts to complete research papers.
Missed the NeurIPS deadline? RLBrew Workshop deadline in less than 15 days. Submit your finished or unfinished work to this RLC workshop.
⚠️ Reminder! Submissions for @RL_Conference's RL beyond Reward Workshop are due May 30 (AoE)!
We are brewing an interesting program and seeking innovative research work in reward-free RL. All papers are welcome, from exploratory abstracts to complete research papers.
It's exciting to think about the capabilities of zero-shot RL as foundation models. Using our work RLZero, you can specify your task as natural language prompts and expect zero-shot policy generation for embodied agents. (1/n)
🤖 Introducing RL Zero 🤖: a new approach to transform language into behavior zero-shot for embodied agents without labeled datasets! RL Zero enables prompt-to-policy generation, and we believe this unlocks new capabilities in scaling up language-conditioned RL, providing an interpretable link between RL agents and humans and achieving true cross-embodiment transfer.
RLZero imagines the trajectory for the language prompt which is used to produce a policy through zero-shot imitation learning. Opens interesting avenues to apply our recent work on zero-shot RL, PSM (arxiv.org/abs/2411.19418) (2/n)
PSM provides an efficient algorithm to learn the basis of the solution space. Any point on the plane is a solution of the Bellman Flow for the MDP. Obtaining the optimal policy for any reward reduces to a simple constrained linear optimization problem. (3/n)
Check out our work for further details.
Paper: arxiv.org/abs/2411.19418
This work couldn’t have been possible without amazing collaborators: @harshit_sikchi @PeterStone_TX and @yayitsamyzhang
@utlarg @lab_midi @texas_robotics @MLFoundations(4/n)