Imitation Learning Hands-on Event Report
Imitation Learning Hands-on Event Report
Date: Monday, September 29, 2025, 14:00-18:00
Venue: Kyoto University Main Campus, International Science Innovation Building 4th Floor, Meeting Room C
Participants: 6 people
⏰Time Schedule
Opening & Environment Setup
Setup of SO-100 robot arm and LeRobot library
Data Collection Practice
Collecting data for imitation learning by operating the robot arm
ACT Training Execution
Executing training with ACT (Action Chunking with Transformers) using collected data
Discussion & Networking
Technical explanation and participant networking utilizing training wait time
Verification & Result Confirmation
The robot actually performs the task with the trained model
Reflection & Closing
Visualization and consideration of attention maps, overall reflection
📊Main Content and Outcomes
KUPAC held its first full-scale hands-on event. Using the SO-100 robot arm and the LeRobot library, participants experienced the entire process of imitation learning (ACT). Participants were able to learn the sequence of steps from data collection to training and verification on the actual machine in a practical manner.
Particularly Impressive Outcomes
- • Confirmed generalization performance beyond expectations, such as success even with workpieces of colors not in the training data
- • Understood the robot's "attention points" through visualization of attention maps
- • All participants completed the practice to the end and succeeded in operation check on the actual machine
- • Active discussion and knowledge sharing among participants from different specialized fields
📸Hands-on Photo Gallery
Data Collection Scene (1)
Data Collection Scene (2)
Hands-on Scene
* Click photo to enlarge
Participants' Voices
"It was very helpful because it would be quite difficult to do it from scratch by myself!"
"It was very educational because it was well prepared."
"Thank you for the easy-to-understand explanation of ACT"
About Future Hands-on Events
We were able to gain a lot of learning and results from this hands-on event. We will continue to hold hands-on events regularly and provide practical learning opportunities for Physical AI.
Details of the next hands-on event will be announced as soon as they are decided.