Max Robotics — Global Service Robot Directory + US Certification

About

Fengyi Zhang is a researcher advancing the frontiers of robot learning and reinforcement learning (RL), with a focus on making autonomous skill acquisition more efficient and robust. Zhang’s work centers on two key areas: developing novel neural network architectures for robotic control and refining the algorithmic foundations of actor-critic methods. A major contribution is the introduction of WAGNN (Weighted Aggregation Graph Neural Network), which enables robots to learn complex skills by structuring their state spaces through graph-based representations. This work, along with SURRL (Structural Unsupervised Representations for Robot Learning), has garnered significant attention, with both papers accumulating 8 and 7 citations respectively, reflecting their impact on the field. Zhang also tackled a critical challenge in RL—the bias-variance trade-off in advantage estimation—by proposing an adaptive mechanism for actor-critic algorithms. This innovation, detailed in a 2023 paper (8 citations), allows critics to dynamically balance variance from sample returns and bias from parameterized value functions, leading to more stable and sample-efficient learning. By addressing fundamental limitations in RL and pioneering graph-based learning for robotics, Zhang’s work is shaping the next generation of autonomous systems capable of mastering intricate tasks without human intervention.