I am a Ph.D. student at Örebro University in Sweden, under the supervision of Martin Magnusson, Todor Stoyanov, and Johannes A. Stork. I am working on the DARKO EU project and affiliated with Wallenberg AI, Autonomous Systems and Software Program (WASP). My research focuses on improving learning efficiency in robotic manipulation, aiming to enable robots to adapt to diverse tasks and move toward general-purpose capabilities with limited supervision. In pursuit of this goal, I focus on Intrinsic Reward-based Exploration, Hierarchical RL, and Robotic Foundation Models.
In 2019, I received my master's degree at the Multimedia Lab in NTNU, Taiwan, focusing on computer vision and domain adaptation under the supervision of Mei-Chen Yeh, and subsequently worked as a Research Associate. Before embarking on my doctoral journey, I worked in the industry as a machine learning engineer.
Working on data-efficient robotic manipulation with intrinsic-reward exploration, hierarchical RL, and robotic foundation models.
Worked on Vision-Language-Action (VLA) models for robotic manipulation.
A hierarchical RL-MPC framework that treats human motion as guidance rather than obstacles. Instead of merely avoiding people, the robot actively aligns with compatible crowd flows.
Keywords: Hierarchical Reinforcement Learning, Crowd Navigation
A continuous spatio-temporal map of dynamics using implicit neural representations, achieving more accurate and efficient motion pattern representation and faster training.
Keywords: Neural Implicit Representation, Spatio-temporal Motion Representation
KEA improves exploration efficiency in sparse reward environments by proactively coordinating exploration strategies between SAC's stochastic policy and novelty-based exploration methods.
Keywords: Intrinsic Reward-based Exploration, Soft Actor-Critic, Reinforcement Learning
We suggest using hierarchical reinforcement learning with parameterized primitives to solve occluded grasping tasks without the need for complex manual controller design.
Keywords: Contact-rich manipulation, Hierarchical Reinforcement Learning
Reviewer for ICML, ICLR, NeurIPS, CoRL, and ICRA.
Örebro University, TUM, Bosch, University of Pisa, EPFL, University of Lincoln, ACT Operations Research
DARKO is innovating agile production robots for efficient and safe intralogistics in warehouses. In this project, I mainly contribute to WP2: 3D Perception and Scene Understanding.
Website
Yufei Zhu*, Shih-Min Yang*, Rishi Hazra*, Kamran Hosseini, Karol Wojtulewicz
We designed a robotic system for precise object manipulation, integrating a perception module, visual reasoning module, and an in-hand perception and control module. Yufei, Rishi, and I applied for this project through WASP. My primary contribution focused on the in-hand perception and control module.
Website