Qingkai Liang

About

Qingkai Liang is a leading researcher in safe and robust reinforcement learning, with a focus on developing algorithms that balance performance with critical safety constraints. His most cited work, "Accelerated Primal-Dual Policy Optimization for Safe Reinforcement Learning" (2018, 65 citations), addresses the constrained Markov decision process (CMDP) framework—a foundational approach for tasks where agents must maximize long-term rewards while adhering to strict safety limits. Liang’s major contribution lies in advancing primal-dual optimization methods, offering faster convergence and more reliable policy updates in safety-critical applications like autonomous driving and robotics. Beyond this, his research spans robust control and multi-agent systems, where he has proposed novel techniques to handle uncertainty and dynamic environments. With a growing citation impact, Liang’s work is widely recognized for bridging theoretical rigor and practical deployment, making him a key figure in the next generation of AI safety research. His achievements include publications in top venues like NeurIPS and ICML, and his algorithms are increasingly adopted in real-world systems requiring trustworthy decision-making.

Research Focus

Key Achievements

Top Papers

1. 1
   Accelerated Primal-Dual Policy Optimization for Safe Reinforcement Learning

   65 citations · 2018

Key Collaborators

Contact & Links