ROS Resources: Documentation | Support | Discussion Forum | Service Status | Q&A answers.ros.org

[TB3] The Delivery Arcade Project (Open Source) - Part1

Delivery Arcade Project - Part 1

This article is sponsored by ROBOTIS

Hi, this is Guerilla-Coders, a team of two undergraduate students in South Korea. We are interested in AI application in tangible world like: 3D printing, or robotics!

Huge thanks to ROBOTIS’ Turtlebot 3 and RAONECH’s AR glasses, we could launch our project “Delivery Arcade”, a platform that supports delivery services in a game-like way, and even allow users to earn money!

By activating Delivery Arcade application on your mobile phone, you can remote control a delivery robot while sharing the pov of the robot from the AR glasses.

So here is what we expect from this project:

Synopsis

There are a few robots waiting to be controlled on the street. Who controls these? Is it being controlled by a delivery agency cloud server? Or is it a self-driving robot?

Here in Anam-campus town’s dorm, Ms. Rilla is sitting on a chair, nothing to do. Rilla is watching YouTube with AR goggles she bought recently. Now that all the videos to watch are gone, so she decided to play the game with goggles. It’s a game where you control a small robot to move objects in time. However, the reality of the game is quite un-real. Can these graphics be available on normal smartphones?

As Ms. Rilla logs into the game, a robot starts moving on the street. The environment that this robot sees (via camera) is transmitted to the goggles as it is, so you can share your view as if you were riding on a robot, or just being the robot itself. Rilla’s game is not in the virtual world, but in the real world.

blurred_sample_image, delivery arcade

We want to create a platform where artificial intelligence and humans work ‘together’ for delivery service.

Autonomous driving is a cutting-edge technology. However, particularly for delivery robots that need to drive with pedestrians require even more advanced technology than self-driving cars. This is because controlled signaling systems exist and there are much more variables and stochasticity than roadways with limited types of obstacles. To develop algorithm to overcome this problem, it is super important to obtain high-quality data to be used for learning, but collecting this data is also quite difficult. One of the reasons for this is because the way a human deliveryman operates and the way a robot works is completely different.

Humans are usually driven by two-wheeled electric motors (figuratively, two legs), but autonomous robots including space rovers are often designed with six wheels for safety.

This is because humans can self-recover when an electric motor gets stuck in a groove on the road surface, but a robot cannot. The point of view of human is in much higher dimension than that of a robot. Thus the robot needs features like additional wheels, but this could easily be neglected if we only considered the human-view issue. Unfortunately, there are tons of complex problems when it comes to collecting optimal data and feature information for autonomous driving including this.

Delivery Arcade is a platform that collects all optimal data in order to fill in the inevitable flaws in current autonomous driving algorithm by handling the problem with human intelligence, rather than artificial intelligence. As mentioned above, this project is quite simple: “controlling a robot remotely through AR goggles”. These ideas are by no means new or novel. However, what sets this project apart from others is that it encourages voluntary participation by designing the user’s robot control process to feel like a game to gather the data.

User pov sample image, delivery arcade

The three most fundamental elements of an AI reinforcement learning model are the environment, behavior, and reward. Surprisingly, the ‘game’ already has all these elements. If the user controls the delivery robot with the sense of labor (I mean, a workload to do with no instant reward), the ‘proper behavior’ → ‘reward’ system for optimal robot driving data is not completed. On the other hand, if the user recognizes it as a game, actions will be taken in the direction of maximizing rewards by avoiding deducting factors such as food being shaken too much and reducing delivery time. This can greatly improve the learning efficiency!

summary:

  • Share the robot’s pov in the AR environment so that the user can control it as it is by himself

  • The robot is basically placed freely on the street like a shared mobility.

  • Users can perform delivery tasks by being assigned a robot in waiting.

  • Users can earn pocket money just like playing games

  • Companies can improve autonomous driving technology by collecting user driving data

In an upcoming post, we are going to reveal the project development process and the finished product as open source!

Thank you!

1 Like