Video: YouTuber’s 3D-printed robot can play entire game of Fetch alone

A YouTuber has unveiled a new AI-powered autonomous robot that can play an entire game of Fetch alone. The robot, interestingly, is also named Fetch by its maker.

This build operates as a strictly offline, closed-loop hardware stack driven by bare-metal Arduino Uno logic. The robot can spot an orange ball and, with no input from a human, go get it and return it back to the owner.

The robot’s 3D-printed components have been put in place with a few screws. It is meant to be friendly in appearance. As soon as an orange ball is in its sights, the 3D-printed arms reach out to take hold of it. Motorised wheels do the work of getting it around. After securing the ball, Fetch will scan the room for its owner. A raised hand is enough for it to make the connection and return the ball.

Autonomous technology helps the robot carry out the work

It is the AI that enables Fetch to tell objects and human gestures apart. The robot has been built using an Arduino UNO Q board with a camera and computer vision. There are two processors with one responsible for analysing images and detecting objects and another controlling the motors and servos of the robot.

Sector 07 trained the computer vision using Edge Impulse. A YOLO-Pro model was used to detect objects within the smaller device. This was trained to detect an orange ball and a human hand, 200 images were taken for each item using the camera on the robot. These were taken in different lighting conditions, floor textures and backgrounds to help the robot identify objects in different environments.

Once trained, the model will create a box around the detected object. Depending on where the object is, the robot will turn its wheels to move in the direction of the ball or find the owners hand.

It will also calculate how far away the robotic arms should be to pick up/let go of the ball.

Fetch comes with small surprises

There are many fun features included in the Fetch. The robot has animated eyes on the front, ears that can move depending on certain situations and voice responses using audio clips created using ElevenLabs.

Throughout the design and build of the robot there were some issues. Sometimes during testing the robot struggled to detect the ball in high lighting conditions. It also struggled to let go of the ball. However these issues were overcome by changing the speed of the motors, grip times and detection settings.

The robot is now completely up to play a game of fetch on its own. The maker has released project as open source. Those interested can access the print files, code, components needed to build the robot, and directions on putting it together at Arduino Project Hub.

​

Source: https://interestingengineering.com/ai-robotics/robot-fetch-youtuber