Xinyi Zhang
Welcome to my homepage! My name is Xinyi Zhang.
I am a Master's student at the
School of Computer Science,
Beijing Institute of Technology (BIT).
I work on reinforcement learning and robot manipulation, with interests in embodied intelligence, real-world RL, and end-to-end robot learning. My goal is to build agents that can explore on their own and keep learning over time, so robots can adapt to messy environments and solve a wide range of tasks with little human supervision.
I work on reinforcement learning and robot manipulation, with interests in embodied intelligence, real-world RL, and end-to-end robot learning. My goal is to build agents that can explore on their own and keep learning over time, so robots can adapt to messy environments and solve a wide range of tasks with little human supervision.
Experience
News
- 2025.12: 👀 We release “Real-world reinforcement learning from suboptimal interventions” for efficient real-world robot manipulation via RL.
- 2025.06: 🎉 Our paper “Training-free generation of temporally consistent rewards from vlm” has been accepted to ICCV 2025.
Publications
* Equal contribution. † Project leader. ✉ Corresponding author.
Real-world reinforcement learning from suboptimal interventions
Yinuo Zhao, Huiqian Jin, Lechun Jiang, Xinyi Zhang, Kun Wu, Pei Ren, Zhiyuan Xu†, Zhengping Che†, Lei Sun, Dapeng Wu, Chi Harold Liu, Jian Tang✉.
arXiv, 2025
SiLRI uses a learnable state-wise Lagrange multiplier to balance imitation and RL under suboptimal expert interventions, cutting the time to 90% success by at least 50% versus HIL-SERL and reaching 100% success on long-horizon tasks.
Training-free generation of temporally consistent rewards from vlm
Yinuo Zhao, Jiale Yuan, Zhiyuan Xu, Xiaoshuai Hao, Xinyi Zhang, Kun Wu, Zhengping Che, Chi Harold Liu✉, Jian Tang✉.
ICCV 2025
T2-VLM is a training-free framework that pairs spatially-aware subgoals with Bayesian tracking to generate temporally consistent VLM rewards for robot manipulation, improving long-horizon control and failure recovery at lower cost.