Robot systems · imitation learning

SEEGIN / liangxianzhong

I work on the engineering layer between a learning algorithm and a robot that actually moves. 我关注学习算法与真正运动起来的机器人之间,那一层常被忽略的系统工程。

I care less about making a model look clever in isolation than about what happens when it enters a system: where the data came from, what the action means, and how failure is recorded. 我不太关心让一个脱离系统的模型显得聪明。我更在意它进入系统以后: 数据从哪来、动作究竟代表什么,以及失败有没有被如实记录。

01Trace the whole path before tuning the model.先把整条链路跑通,再谈调参。
02Keep the evidence beside the claim.结论旁边放证据,数字旁边写口径。
03Leave the failed case in the report.失败案例不删,它定义能力边界。
00

Now近况

2026.08
  • Completed my undergraduate thesis on ACT for household robots and graduated from Xiamen University in 2026.06.刚完成本科毕设《家务机器人 ACT 算法的优化与实现》,2026.06 从厦门大学计算机系毕业。
  • Joining the Institute of Software, Chinese Academy of Sciences in fall 2026, with an interest in systems software and robotics infrastructure.2026 年秋季入学 中国科学院软件研究所,方向偏底层系统软件与机器人系统支撑。
  • Turning 20+ thesis-defense notes into a smaller set of public technical notes.在整理毕设期间写的 20+ 份技术手册,准备把其中一部分公开出来。

Index导览