Robot dog can walk after ONE hour of training as scientists hope it can play fetch in the future

Scientists have imagined a robot dog that can learn to walk in just an hour.

In the video released by the researchers, the four-legged robot first swings its legs in the air and struggles, but after 10 minutes it can take steps, and by the one hour mark it is already walking quite easily, rolling off its feet. back and even walk, being knocked down by a stick by one of the explorers.

Unlike many robots, this robot was not shown in advance what to do in a computer simulation.

Daniyar Hafner, AI researcher at the University CaliforniaBerkeley worked with his colleagues to train a robot using reinforcement learning.

Scroll down for video

According to researchers at the University of California at Berkeley, the robot dog was trained to walk, roll over and overcome obstacles in about an hour. In the photo above, the robot is at the five-minute mark.

This type of machine learning is about taking actions in response to the environment and explores the algorithms used to teach a robot to walk from scratch in the real world.

The team used an algorithm called Dreamer, which uses past experience to build a model of the real world and also allows the robot to make calculations through trial and error.

“The Dreamer algorithm has recently shown great promise for learning from a small number of interactions by planning within a learned model of the world,” the researchers state in their study. paperwhich has not yet been reviewed by experts.

“Learning a model of the world to predict the outcomes of potential actions enables planning in the imagination, reducing the amount of trial and error required in a real environment.”

The researchers used an algorithm called Dreamer, which uses past experience to create a real-world model that the robot can learn from.  In the photo above, the robot is at the 30th minute.

The researchers used an algorithm called Dreamer, which uses past experience to create a real-world model that the robot can learn from. In the photo above, the robot is at the 30th minute.

“Reinforcement learning will be a cornerstone in the future of robot control,” said scientists not involved in the study.  In the photo above, the robot is at the 40th minute.

“Reinforcement learning will be a cornerstone in the future of robot control,” said scientists not involved in the study. In the photo above, the robot is at the 40th minute.

Once a robot has learned to walk, it can also learn to adapt to other, less predictable outcomes, such as when researchers poked it with a stick.

“The problem is that your simulator will never be as accurate as the real world. There will always be aspects of the world that you miss out on,” explains Hafner. MIT Technology Overview.

By the one hour mark, the robot dog pictured above can navigate its environment pretty well, roll over, and more.

By the one hour mark, the robot dog pictured above can navigate its environment pretty well, roll over, and more.

Jonathan Hurst, a professor of robotics at Oregon State University who is not directly involved in the study, told the tech publication that the team’s findings clearly show that “reinforcement learning will be a cornerstone in the future of robot control.”

Even with reinforcement learning, teaching robots to act correctly in the real world is extremely difficult, as engineers must program whether each action is rewarded or not, depending on whether the scientists desire it.

“While the Dreamer shows promising results, training on the equipment for many hours leads to wear and tear on the robots, which may require human intervention or repair,” the researchers state in the study.  In the photo above, the robot overcomes an obstacle.

“While the Dreamer shows promising results, training on the equipment for many hours leads to wear and tear on the robots, which may require human intervention or repair,” the researchers state in the study. In the photo above, the robot overcomes an obstacle.

“The roboticist will have to do this for every task. [or] problem they want the robot to solve,” explains Lerrel Pinto, assistant professor of computer science at New York University, who specializes in robotics and machine learning.

This would amount to a voluminous amount of code and a number of situations that are simply impossible to predict.

The research team names other barriers to this type of technology:

“While the Dreamer shows promising results, training on the equipment for many hours leads to wear and tear on the robots, which may require human intervention or repair,” the study says.

“Additionally, more work is required to learn Dreamer’s limitations and our baseline through longer training.

“Finally, we are looking at solving more complex problems, perhaps by combining the benefits of rapid real-world learning with those of simulators, as an effective direction for future research.”

Hafner says he hopes to teach the robot to obey verbal commands and attach cameras to a dog to give it vision, all of which will allow it to perform more typical canine activities such as playing fetch.

In a separate study, scientists at the German Max Planck Institute for Intelligent Systems (MPI-IS) showed that their robot dog, named Morty, can easily learn to walk using a complex algorithm that includes sensors in its paws.

“As engineers and roboticists, we were looking for an answer by building a robot that has animal-like reflexes and learns from mistakes,” says Felix Ruppert, a former doctoral student in the Dynamic Locomotion Research Group at MPI-IS. statement.

“If an animal stumbles, is it a mistake? Not if it happens once. But if it stumbles a lot, it gives us an idea of ​​how well the robot is walking.”

The robot dog works using a complex algorithm that determines how it learns.

The information from the foot sensors is matched with data from the spinal cord of the machine model, which runs like a program inside the robot’s computer.

The robotic dog learns to walk by constantly comparing set and expected information from sensors, triggering reflex loops, and adapting the way it regulates its movements.

Scientists at the Max Planck Institute for Intelligent Systems in Germany have trained a robot dog known as Morty to walk using algorithms.

Scientists at the Max Planck Institute for Intelligent Systems in Germany have trained a robot dog known as Morty to walk using algorithms.

WHAT IS BOSTON DYNAMICS MINI ROBO DOG?

Boston Dynamics first showed off Spot, the most advanced robot dog ever made, in a video released in November 2017.

The firm best known for its 1.7-meter-tall Atlas humanoid robot has unveiled a new “lighter” version of its Spot robot.

The robot dog was shown running around the yard with the promise that more information would be “coming soon” from the infamous secret firm.

“Spot is a small, four-legged robot that fits comfortably in an office or home,” the company’s website says.

It weighs 25 kg (55 lbs) or 30 kg (66 lbs) including arm.

Spot is all-electric and can run for about 90 minutes on a single charge, depending on what it’s doing, the firm says, boasting, “Spot is the quietest robot we’ve built.”

The Spot was first introduced in 2016, and a previous version of the mini version of the Spot, with the odd retractable neck, was showcased helping around the house.

A previous video from the company shows the robot exiting the firm’s headquarters and entering what appears to be a home.

There, he helps load the dishwasher and carries the trash can.

Also at one point, he collides with a fallen banana peel and plummets, but uses his retractable neck to get back up.

“Spot is one of the quietest robots we have ever built,” the company says, “due to its electric motors.

“It has a variety of sensors, including depth cameras, a solid-state gyroscope (IMU), and proprioception sensors in the limbs.

“These sensors help with navigation and mobile manipulation.

“Spot performs some tasks autonomously, but often uses a human for high-level guidance.”