@article{Han2025TransferrableRobot,

title = {Lessons Learned from the euROBIN Manipulation Skill Versatility Challenge at IROS 2024},

author = {Xudong Han and Haoran Sun and Ning Guo and Sheng Ge and Jia Pan and Fang Wan and Chaoyang Song},

url = {https://msvc-dlrg.github.io/},

doi = {10.1109/RAP.2025.3648290},

year  = {2025},

date = {2025-12-24},

urldate = {2024-12-15},

journal = {IEEE Robotics and Automation Practice},

volume = {1},

pages = {18-22},

abstract = {Robotic automation in high-mix, low-volume settings often relies on structured workcells where fixtures, workpieces, and procedures are well defined, but tasks must be reconfigured quickly across variants. This paper presents the Design and Learning Research Group's solution to the euROBIN Manipulation Skill Versatility Challenge (MSVC) at IROS 2024, in which a standardized electronic task board is operated through a fixed sequence of subtasks. We deliberately adopt a simple, industry-style implementation: each subtask is taught by recording an open-loop Tool Center Point (TCP) and gripper trajectory in the task-board frame, a wrist-mounted depth camera is used once per run to localize the board and a few key features, and a web dashboard exposes the resulting motion primitives as reusable code blocks that can be reordered and reparameterized. With this pipeline, our system completes the benchmark in 28.2 s in the lab (37.2 s on-site), compared to a human baseline of 16.3 s and an average of 83.5 s for previous teams, closing a large part of the performance gap without advanced learning or feedback control. We further demonstrate limited skill transfer by retargeting the same subtask library to a smoke-detector battery replacement scenario required by the competition. Finally, we release our implementation as open source and discuss how competition design and benchmark structure influence the balance between simple engineered solutions and richer sensing and learning.},

note = {Inaugural Issue of IEEE Robotics and Automation Practice},

keywords = {Authorship - Corresponding, Jour - IEEE Robot. Autom. Practice (RA-P)},

pubstate = {published},

tppubtype = {article}

}