Yinsen Jia

What if robots could remember and learn from their own fastest success, even when it came from a “lucky” trial? People often treat efficiency as something to optimize after success. Our new work, Temporal Self-Imitation Learning (TSIL), takes a different view: fastest success itself can be a useful training signal in reinforcement learning. TSIL turns rare fast successes discovered during interaction into two learning signals: adaptive temporal targets that encourage faster completion in a self-paced way, and fast-success self-imitation that helps preserve efficient behaviors before they are forgotten. Across 15 challenging long-horizon manipulation tasks, TSIL improves learning efficiency, task-completion efficiency, and training stability. If training your robot policy feels like endless tuning, give TSIL a shot😄! Grateful to my advisor @Boyuan__Chen for the guidance and support on this work. - Paper: arxiv.org/abs/2606.19752 - Project page: generalroboticslab.com/TSIL - GitHub: github.com/generalrobotic…

@ArithmancyAI Thanks!

@YinsenJ Love the idea of learning from the fastest success. Really clever work.

What if robots could remember and learn from their own fastest success, even when it came from a “lucky” trial? People often treat efficiency as something to optimize after success. Our new work, Temporal Self-Imitation Learning (TSIL), takes a different view: fastest success itself can be a useful training signal in reinforcement learning. TSIL turns rare fast successes discovered during interaction into two learning signals: adaptive temporal targets that encourage faster completion in a self-paced way, and fast-success self-imitation that helps preserve efficient behaviors before they are forgotten. Across 15 challenging long-horizon manipulation tasks, TSIL improves learning efficiency, task-completion efficiency, and training stability. If training your robot policy feels like endless tuning, give TSIL a shot😄! Grateful to my advisor @Boyuan__Chen for the guidance and support on this work. - Paper: arxiv.org/abs/2606.19752 - Project page: generalroboticslab.com/TSIL - GitHub: github.com/generalrobotic…

@woodtechwill The repo license has been updated to Apache-2.0. Feel free to try it out :)

@YinsenJ CC BY-NC-ND. So you can look but not touch. Typical.

A related direction from our earlier work: once a policy can solve a task, can time become a control signal for robot behavior at inference time⏲️? We study time as a control dimension for inference-time policy adaptation, enabling diverse, adaptive, robust, and timely behaviors in changing environments without retraining to meet new timing requirements. Check out the links below if you are interested in time-aware policy inference and adaptation! - Paper: arxiv.org/abs/2511.07654 - Video: youtube.com/watch?v=l3BG9S… - Project page: generalroboticslab.com/TimeAwarePolicy

Want a quick intuition for how TSIL works? We made a 5-min Colab demo! In a tiny maze, you can see how a rare fast success becomes reusable supervision: TSIL updates an adaptive temporal target and reuses efficient behavior through self-imitation. Try the demo here: colab.research.google.com/github/general…

@chichengcc This is so cool! Congrats Cheng :)

mm level precision beyond actuator limits, so much torque that you need to manage thermals. Owning the whole stack from HW to AI is the only way 🦾

We present ClutterGen that tackles two important research problems in robotics: 1) how to generate physically compliant cluttered scene for robot learning? 2) how to stably place objects instead of the current "pick and drop" paradigm? Our key idea is to formulate physically compliant cluttered scene generation as a sequential object placement problem, such that we can look at scene generation as a reinforcement learning problem. The learned environment generator can be easily applied to solve truly "pick and place" tasks. Our robot can figure out how to place an object in a stable manner into a cluttered environment. Please check out our project website with video, code, and paper to learn more! generalroboticslab.com/ClutterGen Great work done w/ @YinsenJ at Duke General Robotics Lab (generalroboticslab.com)