Author: Yu Hang He, Email: hey6@unlv.nevada.edu
Date Last Modified: 07/23/2018

This week, I had the opportunity to visit Chungnam National University in South Korea. The CNU campus is about 45 minutes walk away from the KAIST campus. Similar to KAIST, CNU campus is another beautiful campus with scenic walkways and plentiful of art exhibits along the path. I learned that there are many exchange students in CNU because they have a language school for learning Korean attached to the university.

Right between KAIST and CNU is a popular shopping and dining district called Gungdong. The district is popular for both KAIST and CNU students and professors because it is right around the corner. The district have many different styles of restaurants available from traditional Korean foods to Japanese and Western foods. There are also many different stores and entertainments. I went to restaurants in Gungdong on several occasions. The foods were delicious.

I think it is important to talk about Korean's work culture. During my time in the Hubo Lab, I was able to gain insight into Korean's work culture. Korean have a much stricter working schedule than United States and probably most western society. A typical day start at 9:00 AM and end at around 10:00 PM in the Hubo Lab.

Last week, I tried to implement the inverse kinematic function into my mocap project. However, after some trial and error, I realized that the available inverse kinematic are not suitable for for this project. After consideration, I decided to implement the inverse kinematic function outlined in this paper by Sinichiro Nakaoka

Currently, the new inverse kinematic function takes three positional inputs at shoulder, elbow, and wrist to calculate joint position at shoulder pitch, yaw, roll and elbow. The new inverse kinematic function are much better suited to imitate motion captured through mocap systems. Furthermore, I can attach additional markers in future experiments to calculate wrist pitch and yaw positions.

Additionally, I implement the filtering functions directly into the Hubo 2's PODO ALPrograms. The new ALProgram will directly parses the positional data captured through mocap system and filter and convert them into joint positional data. The filter implemented in this project is locally weighted scatterplot smoothing method, which I found to be most robust and accurate in filtering both the trajectory and joint data.

I was able to combine everything into a program on Hubo2 that can be operated through GUI to play recorded mocap data for the right arm. The next step will be to limit the Hubo 2's motion within workspace and reduce the joint velocity and acceleration when the motion of human exceeds the speed of Hubo 2.