Kibo RPC Code — 6th Challenge

Code from our team's participation in the 6th Kibo Robot Programming Challenge (Kibo-RPC), organized by JAXA and NASA. We finished 9th nationally.

What is Kibo-RPC?
The Kibo Robot Programming Challenge is an annual competition in which student teams program Astrobee, a free-flying robot aboard the International Space Station (ISS), to autonomously complete a series of tasks. Teams write code that runs on the actual ISS hardware after passing through simulation rounds.

---

Repository Structure

kibo-rpc-code/
├── code-using-opencv/   # Java — robot control using OpenCV for image detection
├── code-using-yolo/     # Java — robot control using a custom-trained YOLO model
└── yolo code/           # Python — model training scripts used to produce the YOLO weights

How the two YOLO folders relate:
The Python code in yolo code/ was used to train a custom YOLO object detection model. The resulting weights were then bundled into the Java project in code-using-yolo/, which runs on the Astrobee robot during the competition.

---

Approaches

OpenCV (code-using-opencv/)

Classical computer vision approach written in Java. Handles target detection using OpenCV image processing techniques within the Kibo-RPC SDK.

YOLO (code-using-yolo/ + yolo code/)

A two-part implementation: a Python training pipeline to produce a custom-trained YOLO model, and a Java implementation that runs inference on the robot via the Kibo-RPC SDK. The YOLO approach yielded more robust detection under varying ISS lighting conditions.

Detection output (YOLO validation batch):

Simulator (RViz) running the robot navigation code:

---

Results

Finished 9th place nationally in the 6th Kibo-RPC.