Find Something You Can't Do: Agentic Real-World Reinforcement Learning for Self-Improving VLA Models
arXiv 2026
Enables VLA self-improvement through autonomous real-world practice, adaptive task selection, and self-evaluation.
Publications
* equal contribution
2026
HARBOR: A Harness Framework for Agentic Robot Reinforcement Learning
CoRL 2026
Automates the end-to-end robot RL workflow, from package installation to policy tuning.
Scalable Multi-Task Data Generation via Reinforcement Learning for Language-Conditioned Bimanual Dexterous Manipulation
IROS 2026
Generates multi-task demonstrations for language-conditioned bimanual manipulation via RL.
Nautilus: From One Prompt to Plug-and-Play Robot Learning
NeurIPS 2026
Turns a single prompt into reproduction, evaluation, fine-tuning, and deployment workflows.
1 more from 2026Show fewer
SE(3)-PoseFlow: Estimating 6D Pose Distributions for Uncertainty-Aware Robotic Manipulation
ICRA 2026
Estimates full 6D pose distributions via flow matching on SE(3) for uncertainty-aware manipulation.
2025
Morphologically Symmetric Reinforcement Learning for Ambidextrous Bimanual Manipulation
CoRL 2025
Exploits morphological symmetry to learn ambidextrous bimanual manipulation.
DIME: Diffusion-Based Maximum Entropy Reinforcement Learning
ICML 2025
Trains diffusion policies with maximum-entropy RL, reaching state-of-the-art returns.
2024
Learning Multimodal Behaviors from Scratch with Diffusion Policy Gradient
NeurIPS 2024
Learns multimodal diffusion policies from scratch, without demonstrations.
Reconciling Reality through Simulation: A Real-to-Sim-to-Real Approach for Robust Manipulation
RSS 2024
Robustifies real-world policies by fine-tuning them in digital twins built on the fly.
2023
Parallel Q-Learning: Scaling Off-policy Reinforcement Learning under Massively Parallel Simulation
ICML 2023
Scales off-policy RL to 10,000+ parallel environments.
2 more from 2023Show fewer
Social Learning for Sequential Driving Dilemmas
Games 2023
Identifies when social dilemmas arise in autonomous vehicles’ sequential decisions.
Stationary Deep Reinforcement Learning with Quantum K-spin Hamiltonian Equation
ICLR Workshop 2023
Adds a quantum K-spin Hamiltonian regularizer that steers policies toward better optima.
2022
Homomorphic Matrix Completion
NeurIPS 2022
Achieves exact matrix completion under differential privacy, at the cost of more samples.
1 more from 2022Show fewer
Social Learning In Markov Games: Empowering Autonomous Driving
IEEE IV 2022
Studies how autonomous vehicles form social norms through RL in Markov games.
2021
Show 2 papersShow fewer
FinRL-Podracer: High Performance and Scalable Deep Reinforcement Learning for Quantitative Finance
ICAIF 2021
Trains a NASDAQ-100 trading agent in 10 minutes on 80 A100 GPUs.
ElegantRL-Podracer: Scalable and Elastic Library for Cloud-native Deep Reinforcement Learning
NeurIPS Workshop 2021
Scales cloud-native DRL training across millions of GPU cores.
Open Source Projects
FinRL: Financial Reinforcement Learning
GitHub16.2k
Brings deep reinforcement learning to quantitative finance and automated trading.
ElegantRL “小雅”: Massively Parallel Library for Cloud-native Deep Reinforcement Learning
GitHub4.4k
Runs massively parallel DRL training on cloud-native infrastructure.
As a leader of this project, I have been contributing to
- develop a series of large-scale training frameworks,
- implemente SOTA algorithms and techniques,
- build the documentation website.
Tutorial blogs (6)
Starting from Mar. 2021, I started to write tutorial blogs for the community,
- ElegantRL: Much More Stable Deep Reinforcement Learning Algorithms than Stable-Baseline3MLearning.ai, Mar. 3, 2022
- ElegantRL-Podracer: A Scalable and Elastic Library for Cloud-Native Deep Reinforcement LearningTowards data science, Dec. 11, 2021
- ElegantRL: Mastering PPO AlgorithmsTowards data science, May. 3, 2021
- ElegantRL Demo: Stock Trading Using DDPG (Part II)MLearning.ai, Apr. 19, 2021
- ElegantRL Demo: Stock Trading Using DDPG (Part I)MLearning.ai, Mar. 28, 2021
- ElegantRL-Helloworld: A Lightweight and Stable Deep Reinforcement Learning LibraryTowards data science, Mar. 4, 2021
Book Chapter
High-performance Tensor Decompositions for Compressing and Accelerating Deep Neural Networks
Elsevier 2021
Compresses and accelerates neural networks with high-performance tensor decompositions.