Zhiwei Steven Wu

Papers

Total Citations

H-Index

About

Zhiwei Steven Wu is a pioneering researcher in safe reinforcement learning (RL) and robotic manipulation, whose work bridges the gap between theoretical guarantees and real-world deployment. His most impactful contribution, "Constrained Variational Policy Optimization for Safe Reinforcement Learning" (2022, 16 citations), fundamentally addresses the instability and lack of optimality guarantees in primal-dual safe RL methods, offering a principled framework for learning policies that satisfy safety constraints—a critical advancement for deploying AI in safety-critical applications like autonomous driving and healthcare. In robotics, Wu's "A Back-Drivable Rotational Force Actuator for Adaptive Grasping" (2023, 5 citations) introduces a compact rotary series elastic actuator (RSEA) with an innovative arc groove design, enabling adaptive, compliant grasping in robotic hands. His work "Learning Shared Safety Constraints from Multi-task Demonstrations" (2023, 3 citations) tackles the challenge of transferring safety knowledge across tasks, proposing that agents can learn common constraints—like "don't break plates"—from diverse demonstrations, reducing manual specification. Wu's research uniquely combines theoretical rigor with practical hardware innovation, making him a leading voice in creating AI systems that are both capable and safe.