Zixing Lei

Zixing Lei is a researcher at Shanghai Jiao Tong University (SJTU) and affiliated with the Zhongguancun Academy, whose work focuses on agentic systems and vision-language-action (VLA) model post-training for embodied robotics. Lei is the lead author of the 2026 paper "Towards Long-horizon Embodied Agents with Tool-Aligned Vision-Language-Action Models" (arXiv:2605.13119), which introduces TAPT (Tool-Aligned Post-Training), a framework that trains specialized VLA tools for bounded subtasks and coordinates them through a high-level vision-language model agent to tackle long-horizon robotic manipulation. The work demonstrates marked improvements on benchmarks including LIBERO-Long and RoboTwin by decoupling high-level temporal planning from local physical execution across a family of tool-specialized VLA models.