Shengjie Wang | 王圣杰

I am a PhD student in Computer Science at Institute for Interdisciplinary Information Sciences (IIIS), Tsinghua University , working with Prof. Yang Gao. Previously, I was a Master's student in Department of Automation at Tsinghua University , advised by Prof. Tao Zhang(IEEE Fellow & IET Fellow, Head of Department). I completed my Bachelor's in Robotics at BIT, supervised by Prof. Qing Shi (IEEE Senior Member) and Prof. Toshio Fukuda (2020 IEEE President).

Email  /  CV  /  Google Scholar  /  Twitter  /  Github /  WeChat

My research interests lie in the intersection of Robotics and Reinforcement Learning. I care about efficient, stable and safe robot performance in the unstructured open world.

Furthermore, we are passionate about releasing some interesting robotic environments, such as Bi-DexHands and SpaceRobotEnv. Hope everyone enjoys our work!

We propose Semantic Keypoint Imitation Learning (SKIL), a framework which automatically obtains semantic keypoints with the help of vision foundation models, and forms the descriptor of semantic keypoints that enables efficient imitation learning of complex robotic tasks with significantly lower sample complexity.

RoboMetaverse is founded by a team of robotics enthusiasts, dedicated to creating a highly realistic embodied intelligence simulation platform and a cost-effective embodied intelligence hardware platform, aiming to contribute to the realization of the AGI (Artificial General Intelligence) era in embodied intelligence.

We present a learning-based catching strategy, which can catch diverse objects of daily life with dexterous hands. The learned policies show strong zero-shot transfer performance on unseen objects.

We propose Robotic Manipulation through Spatial Constraints of Parts (CoPa), a novel framework that incorporates common sense knowledge embedded within foundation vision-language models (VLMs), such as GPT-4V, into the low-level robotic manipulation tasks.

Introducing EfficientZero V2, a general framework designed for sample-efficient RL algorithms, which outperforms the SotA methods and DreamerV3 across diverse domains.

we present the i-Octree, a dynamic octree data structure that supports both fast nearest neighbor search and real-time dynamic updates, outperforming contemporary state-of-the-art approaches by achieving, on average, a 19% reduction in runtime on realworld open datasets.

Our approach enables the system's state to reach an equilibrium point within an optimal time and maintain stability there- after, referred to as "optimal-time stability". To achieve this, we integrate the optimization method into the Actor-Critic framework, resulting in the development of the Adaptive Lyapunov-based Actor-Critic (ALAC) algorithm.

Our algorithm (ESB-CPO) improves upon the trade-off between reducing constraint violations and improving expected returns in Safe Reinforcement Learning.

We propose a novel Learning-based Adaptive Compliance (LAC) algorithm to improve the efficiency and adaptability of symmetric bi-manual manipulation.

We propose the Intrinsically Motivated Adversarial Policy (IMAP) for efficient black-box evasion attacks in single- and multi-agent environments without any knowledge of the victim policy.

We developed a small-sized quadruped robotic rat (SQuRo), which includes four limbs and one flexible spine. On the basis of the extracted key movement joints, SQuRo was subtly designed with a relatively elongated slim body (aspect ratio: 3.42) and smaller weight (220 g) compared with quadruped robots of the same scale.

We propose a novel algorithm, EfficientLPT, to facilitate RL-based methods to improve planning accuracy efficiently. Our core contributions are constructing a mixed policy with prior knowledge guidance and introducing infinite norm to build a more reasonable reward function.

We propose a bimanual dexterous manipulation benchmark (Bi-DexHands) according to literature from cognitive science for comprehensive reinforcement learning research.

We developed a model-free Hierarchical Decoupling Optimization (HDO) algorithm to realize 6D-pose multi-target trajectory planning for the free-floating space robot.

Template stolen from Jon Barron.
Last updated: Oct 15, 2023