Xubo Lyu

About

Xubo Lyu is a researcher advancing the frontiers of reinforcement learning (RL) for complex, data-intensive robotic systems. His work bridges the critical gap between model-free and model-based approaches, aiming to make RL more sample-efficient and practical for real-world control. In his highly cited paper "TTR-Based Reward for Reinforcement Learning with Implicit Model Priors" (2020, 4 citations), Lyu introduced a novel reward-shaping technique that leverages implicit model priors to guide exploration, significantly improving data efficiency without sacrificing the flexibility of model-free methods. Building on this, his "MBB: Model-Based Baseline for Efficient Reinforcement Learning" (2020, 3 citations) proposed a hybrid framework that combines the data-efficiency of optimal control with the adaptability of model-free RL, offering a powerful baseline for robotic learning tasks. More recently, Lyu has tackled the challenge of multi-agent coordination with "Asynchronous, Option-Based Multi-Agent Policy Gradient: A Conditional Reasoning Approach" (2023, 2 citations), where he developed a novel policy gradient method that enables agents to reason about conditional dependencies and execute asynchronous, high-level actions. His work is shaping the next generation of algorithms for autonomous systems, from single-robot control to cooperative multi-agent teams.

Research Focus

Key Achievements

Top Papers

1. 1
   TTR-Based Reward for Reinforcement Learning with Implicit Model Priors

   4 citations · 2020
2. 2
   MBB: Model-Based Baseline for Efficient Reinforcement Learning.

   3 citations · 2020
3. 3
   Asynchronous, Option-Based Multi-Agent Policy Gradient: A Conditional Reasoning Approach

   2 citations · 2023

Key Collaborators

Contact & Links