Which machine learning control method works best for robot navigation: Audio, Image, Human Pose Estimation?

I believe that Image Classification for robot navigation is the most efficient technique for controlling the robot because you are able to train on a large dataset of images and is clearer in terms of detecting compared to sound or human pose.

I researched in the field of Machine Learning and Robotics. For Machine Learning, I learned how to use Teachable Machine to classify sound, images, and human poses. Based on this detection, I was able to control a robot built using the ELEGOO kit.

Cotrolled - robot, test maze, sensors, camera

Manipulated - classification method, code

Responding - which classification method is best used for robot navigation?

1. Build robot using ELEGOO kit.

2. Build machine learning models on Teachable Machine (sound, image, human pose)

3. Configure electric circuit (hardware) for testing using LED lights.

4. Combine classification of instance with a robot movement (For example: Saying "GO" will make the robot move forward)

5. Test the three different methods and navigate the robot through a maze.

6. Record results

The image classification method worked the best.

Analysis was taken in the form of a table that records time of the robot taken to navigate through a maze.

In conclusion, image classification worked better than sound and human pose classification methods. I learned that this is mainly due to the misrepresentation that can be detected in estimating a person's body movements or speech.

Code - Arduino Uno

Machine Learning - Teachable Machine

Robot - ELEGOO kit

Camera - Laptop Camera

Sources of error:

- accuracy of model

- test track (maze)

- code bugs

- response time of robot

Teachable Machine

N/A