@svlevine: RL can help collect good data, whether you're doing BC or RL. It can also stack two cubes on top of another cube. Fun p…
Summary
RL can assist robot data collection during demonstrations through a technique called Data Inoculation, improving data quality for both behavior cloning and RL, as shown in a new paper on stacking cubes.
View Cached Full Text
Cached at: 11/03/26, 06:20 AM
RL can help collect good data, whether you’re doing BC or RL. It can also stack two cubes on top of another cube.
Fun paper with @qiyang_li @zhiyuan_zhou_ @CharlesXu0124 on how RL can assist with robot data collection!
Paul Zhou (@zhiyuan_zhou_): We always talk about models in robot learning, yet data quality is just as important (if not more!). How do we collect good data?
It turns out RL can help you during data collection! If you’re collecting demos, ✨Data Inoculation✨ is a simple addition to collect better data!
Similar Articles
@svlevine: How do we run RL with real-time chunking (RTC)? In this work we figured out how to use a small RL policy with a large r…
A research collaboration introduces ARLI, a method to enable RL fine-tuning of Vision-Language-Action models with asynchronous inference using real-time chunking, improving robot behaviors by allowing the RL policy to observe more recent images.
DataFlex-RL: An Evaluation Platform for RLVR Data Policies
DataFlex-RL is an evaluation platform for RLVR data policies that finds uniform sampling performs comparably to adaptive methods across mathematics, logic, and science benchmarks.
@dair_ai: Great paper from Salesforce AI Research on RL for multi-turn tool use. The finding is that you want to train the one ca…
A paper from Salesforce AI Research introduces Critical-State RL, a method to identify and train the most impactful model call in multi-turn tool use trajectories, improving accuracy on the BFCL v4 benchmark by up to 14 points.
From RLVR to RLSVR (GitHub Repo)
Introduces RLSVR, a task-transformation paradigm that extends RLVR to open-ended tasks via self-verifiable rewards from self-play games, instantiated in SpyRL and Vision-Zero. It improves LLM performance on summarization, creative writing, and math reasoning.
@raydistributed: RollArt is an impressive example of disaggregation in large-scale RL. https://cse.ust.hk/~weiwa/papers/rollart-osdi26.p…
RollArt presents a disaggregated architecture for large-scale reinforcement learning, demonstrating significant improvements in efficiency and scalability.