Wenli Xiao

I am a first-year Master (MS, Robotics) student at CMU Robotics Intitute. I am currently working in LeCAR Lab, co-advised by Prof. Guanya Shi and Prof. John Dolan.

Email  /  Google Scholar  /  Twitter  /  Github  /  Photography

My ultimate goal is to build deployable robotic systems and facilitate robots working in our daily lives!
My research focuses on using data-driven approaches and combining large-scale (Reinforcement) Learning-based methods with control-theoretical frameworks to tackle: 1) safety, 2) agility, and 3) adaptivity, to make robots more deployable. I am passionate about real robots and have extensive experience with 1/10 Racing Cars, quadrupedal robots, and Humanoids! (*equal contribution)

TL;DR: H2O enables real-time whole-body teleoperation of a full-sized humanoid to perform tasks like pick and place, walking, kicking, boxing, etc.

Legged robots navigating cluttered environments must be jointly agile for efficient task execution and safe to avoid collisions with obstacles or humans. Existing studies either develop conservative controllers (< 1.0 m/s) to ensure safety, or focus on agility without considering potentially fatal collisions. This paper introduces Agile But Safe (ABS), a learning-based control framework that enables agile and collision-free locomotion for quadrupedal robots.

This paper jointly tackles policy adaptation and safe reinforcement learning with safety guarantees. Comprehensive experiments on (1) classic control problems (Inverted Pendulum), (2) simulation benchmarks (Safety Gym), and (3) a real-world agile robotics platform (RC Car) demonstrate great superiority of SafeDPA in both safety and task performance, over state-of-the-art baselines.

We propose the Model-based Dynamic Shielding (MBDS) framework to address the safety challenges in Multi-Agent Reinforcement Learning (MARL), while providing formal guarantees through the application of Linear Temporal Logic (LTL) in its construction.

This paper presents an adaptive client sampling algorithm that minimizes wall-clock convergence time in Federated Learning by addressing system and statistical heterogeneity, resulting in up to 73% less time spent compared to baselines (weighted sampling, uniform sampling, .etc).