Max Robotics — Global Service Robot Directory + US Certification

About

Fangzhou Xiong is a leading researcher in robotics and reinforcement learning, whose work tackles two of the field’s most persistent challenges: catastrophic forgetting in continual learning and the curse of dimensionality in high-dimensional state spaces. Xiong’s most influential contribution, “Guided Policy Search for Sequential Multitask Learning” (44 citations), advanced the practical application of guided policy search (GPS) algorithms, enabling robots to learn multiple tasks sequentially without succumbing to local optima or requiring prohibitive real-time sample collection. Building on this foundation, Xiong developed the Weighted Aggregation Graph Neural Network (WAGNN) for robot skill learning, a novel architecture that leverages graph neural networks to extract structured policies from complex, continuous control domains. A central theme across Xiong’s work is overcoming catastrophic forgetting—a critical barrier to lifelong robotic learning. Through state primitive learning and encoding primitives generation policy learning, Xiong demonstrated that robots can acquire new manipulation skills without overwriting previously learned knowledge, even when tasks are presented sequentially without stored past data. With over 70 total citations and a consistent focus on scalable, memory-efficient learning, Xiong’s research is shaping the future of autonomous, adaptable robotic systems capable of mastering diverse real-world tasks.