|
Learn about the tech behind the story
How would a Dog-Sitter 3000 robot work in the real world? JoJo uses generative artificial intelligence (AI) to make decisions when watching his dog, Doddy. At the Dog-Sitter 3000 factory, a robot prototype is trained on data, such as videos of dogs following commands or images of animals safe for a dog to meet. A teacher would show the robot how to perform dog care tasks, such as playing fetch with a dog. Then under careful human supervision in a home test environment, the robot would practice caring for a dog. When the robot prototype is ready, it would be manufactured, tested for quality, and shipped to consumers like Josephine and Joseph. Using generative AI, robots like JoJo can continue on their own to improve their skills to become a dog’s best friend. Read on to learn real-world technology words and terms, and then take a deep dive into the technology used in JoJo’s world. |
|
Technology words and terms to know
Artificial intelligence: AI is a computer or machine that can think like a human. AI can learn from data and experience. It can also solve problems and make decisions. AI can learn from its mistakes to increase its accuracy. Artificial intelligence algorithms: AI algorithms are like a set of instructions or rules that tell a computer how to solve problems, make decisions, and perform tasks. Artificial intelligence diffusion policy: An AI diffusion policy is like a guide that helps robots learn new skills by practicing and figuring out the best way to do things. This generative AI technique helps robots learn how to perform tasks autonomously. The AI diffusion policy generates the robot’s behavior, making it possible to perform complex tasks. Autonomous: When robots can perform a task without human help, they can perform autonomously. Generative artificial intelligence: Generative AI works by using complex algorithms and neural network models to create new content or generate behavior in robots. Generative AI can create new content such as text, images, video, music, and more. In robotics, generative AI can help a robot to perform new tasks (see AI diffusion policy). This technology can also help robots communicate more naturally with humans. Neural network models: Inspired by how the human brain works, a computer neural network model has thousands or even millions of simple processing nodes (also called artificial neurons). These models have layers of interconnected nodes that process information and generate new outputs based on what they have learned. Each connection has a weight that adjusts as the network learns from data. During training, the network learns to recognize patterns and features in the data. Once trained, the neural network can generate new content by sampling from the learned patterns. Vector database: A vector database stores and manages vectors, which are lists of numbers that represent data in a way that AI algorithms can understand and process. For example, an image can be converted into a vector where each number represents a specific feature of the image, such as color, shape, or texture. This makes it easier for AI to search, analyze, and find patterns in the data. |
|
|
How technology is used in JoJo’s world
What is really happening to JoJo while he’s dreaming in the story?
When JoJo finds the image of the jackalope (the jackrabbit with antelope horns), he is unable to determine if Binny the bunny is safe to interact with Doddy the dog. This leads to JoJo’s controller throwing an unknown error exception.
To handle the error in real life, the home robot’s controller might work with an exception handler to clear the error and restore the controller to a normal state. This would allow the robot to continue operating without restarting. In the story, we chose to clear the error by having JoJo’s controller do a soft restart. JoJo does not lose any previous knowledge or skills. Instead, special software running during the restart adjusts his AI database to allow Binny the stuffed bunny to interact with Doddy. A restart is a simple technology fix that kids may have already encountered when doing schoolwork or playing games on computers. To go deeper, in our imaginary world, Dog-Sitter 3000 robots may encounter situations where the robot has no relevant prior experience data in its AI vector database, such as identifying a jackalope. To handle these cases, the robot might use software to run a “dream” program on a separate processor during the restart. This would allow the robot to create a dream scenario using prior experience data and dynamically generate new guidance data for future use. In this case, JoJo creates a dream experience where he continuously monitors his dog even when confronted with a worst-case scenario of a rogue rabbit. The simulated experience is based on actual experience data, including pictures of fields and rabbits, and known engine performance specifications for driving over rugged terrain like grass. The dream produces new guidance data in his AI vector database just as a real-world experience would. When JoJo wakes up, he is not concerned about Binny because he has the needed experience to handle wild bunnies in the future. How would a robot like JoJo use generative AI to learn how to take care of dogs?
Researchers are now using large behavioral models combined with generative AI to train robots how to learn and navigate their surrounding environment, according to MIT Technology Review.
For example, researchers at the Toyota Research Institute start the learning process with a human teacher manually controlling the robot to demonstrate a new set of skills, combined with a text description of the goal. During this phase, the generative AI diffusion policy learns how to do the new task. After the demonstration, the robot continues to learn autonomously how to perform the new behavior through trial and error. The AI diffusion policy generates the robot’s behavior, making it possible to perform complex tasks. In our fictional world, JoJo would learn how to give a treat to a dog at the Dog-Sitter 3000 factory. The teacher might show JoJo how to use his end effector gripper to open the treat compartment, pull out a snack, and toss it to the dog. Through practice, JoJo would learn the right order of movements to give treats and when to reward a dog. This factory training provides JoJo with the needed experience to generalize this knowledge for use in a home environment, where he can continue to improve his skills. |
