Discovering the Magic of AI: A Fun Adventure for Kids!

In the fun and exciting world of the FIRST Tech Challenge 2023, I went on a cool adventure to learn about Artificial Intelligence (AI). It was like a big competition where we built robots to do awesome things.

Beginning

At first, I didn’t know much about AI, but the challenge encouraged me to explore and understand it better. AI is like teaching machines to be smart and do things on their own. Think of it like a robot learning to solve puzzles or make decisions by itself.

Exploration

Working with my friends on our robot project was super fun. We had to use AI ideas to make our robot smart and do cool tasks. It was like teaching a robot to be a superhero!

The best part was that we didn’t just learn from books — we got to try things out and solve real problems. Every challenge we faced helped us understand AI better. And you know what? Even if things didn’t go perfectly, we learned from our mistakes and made our robot better

Challenge

We had an exciting challenge in our game where our robot had to find and recognize a special object during the Autonomous period. To make this happen, we needed to teach our robot how to ‘see’ and identify the objects. However, we faced several challenges while creating the algorithm for this task. Overcoming these challenges became a crucial part of our journey in making our robot smart and capable.

Approach

1. We gathered lot of data on our custom object by capturing the images using the camera mounted in our Robot
2. We annotated and labelled the gathered data using the tool Makesense.ai website, it was very easy to use
3. We kept adding more data to make our Robot algorithm really good.
4. Once the data is ready , we categorized the data in to 3 groups — Train, Validate & Test
5. Out of total images we collected we used 70% for Train, 20% for Validate and 10% for Test
6. Based on our mentors recommendations we tried two algorithms YOLO and SSD.
7. Using Train data we trained our Model
8. After Training the model, we used the Validate data to check whether our model is performing good or bad
9. Once we got desired results we used another set of images in “Test” category to evaluate the models performance.

Annotation & Training

Some of the Annotated images are shown below — We labeled two characters Man & Turtle

Training & Outcome

Inference & Outcome

After training the model, we ran the test with new set of images and BINGO it did detect the objects but with varying confidence level. We are still fine tuning our model for this year.

Thank You

This is my [Srishti Sakthivel] FTC journey so far and I have really enjoyed it. I am really proud of myself for being part of the reason this team has went to worlds and for doing all the things I have done while I am in this team and for this team as well.

I’m [Sakthivel Ganesamoorthy] delighted to introduce my child to some exciting trends at this young age.

Support our team Rapid Robots — Team 10477 [Novi — Michigan] by following us on X.com [@RapidRobotsNovi] Insta & Facebook.