I focus on embodied manipulation, and I am currently on a gap year. I am at Xi'an Jiaotong-Liverpool University, advised by Prof. Yong Yue and Prof. Yaran Chen, and I work with Prof. Shangke Lyu at the School of Robotics and Automation, Nanjing University. I previously interned at Westlake Robotics.
One problem in two halves — representing the world, and acting on that representation.
An embodied system that generalizes robustly enough to complete long-horizon, complex manipulation tasks in open environments.
I model how an embodied agent understands and represents the world, and build systems that act in a manner consistent with those internal representations.
Vision-Language-Action policies break down on long-horizon tasks in open environments because they hold no stable representation of task-relevant state, temporal dependency, or transition condition — capping robustness, generalization and success rate at once. My work introduces stronger state modeling and decision-support mechanisms for VLA.
Selected work on embodied manipulation and world representation. All publications →
IROS 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems
A training-free retrofit that gives frozen VLA policies history awareness by caching intermediate-layer prefix K/V states — no parameter change, no longer inputs. +4.0 points on LIBERO-Long with π0.5, and up to +14.2% full-task success on a real Realman RM-65B.
IROS 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems
Research groups and industry teams. Full CV →
Happy to hear about embodied AI, world models and reinforcement learning — research, collaboration, or PhD opportunities.
luciojunsun@gmail.com ↗WeChat · qw2667180048