Xiangkun He

About

Xiangkun He is a robotics researcher whose work sits at the intersection of reinforcement learning, multi-agent systems, and human-robot interaction. His primary contributions lie in developing robust, sample-efficient control algorithms for robots operating in challenging environments—from adversarial and sparse-reward settings to heterogeneous multi-robot teams. He is best known for his work on robust goal-conditioned reinforcement learning, which addresses the critical problem of learning complex end-to-end controllers from high-dimensional sensory data under uncertainty. His 2023 paper on this topic has garnered 27 citations, reflecting its impact on the field. He has also advanced personalized robotics through constrained multi-objective reinforcement learning (17 citations) and pioneered transformer-based multi-agent reinforcement learning to enable combinatorial generalization in heterogeneous robot teams—a key step toward scalable, real-world multi-robot cooperation. He’s recent work on transformer architectures for multi-robot systems, published in 2024, is already gaining attention. Through these contributions, He is shaping how robots learn to adapt, cooperate, and serve humans in complex, dynamic environments.

Research Focus

Key Achievements

Top Papers

1. 1
   Robotic Control in Adversarial and Sparse Reward Environments: A Robust Goal-Conditioned Reinforcement Learning Approach

   27 citations · 2023
2. 2
   Personalized robotic control via constrained multi-objective reinforcement learning

   17 citations · 2023
3. 3
   Transformer-based Multi-Agent Reinforcement Learning for Generalization of Heterogeneous Multi-Robot Cooperation

   4 citations · 2024
4. 4
   Robotic Control in Adversarial and Sparse Reward Environments: A Robust Goal-Conditioned Reinforcement Learning Approach

   2 citations · 2023

Key Collaborators

Contact & Links