US research increases robot skills and awareness
A team of scientists at UC Berkely has developed a robot dog with the ability to teach itself to walk within an hour, according to The Science Times. This is a different approach to the more conventional training of robots in simulation environments.
In what is known as ‘reinforcement learning’, the robot takes its first steps after around 10 minutes and is a competent walker after an hour. This trial-and-error approach involves a lot of stumbles and falls before learning from experience. Through reinforcement training, engineers program when actions will result in rewards based on specific objectives.
Researcher Danijar Hafner believes that mere simulation is not sufficient to prepare robots for the complexities of the real world. In a paper published in Arxiv, the team introduces the Dreamer algorithm. This builds a model of the real world based on past experiences and enables the ability to predict future outcomes through experimentation in the real world.
The robot is not only learning like a newborn but is also watching YouTube videos to see what it can pick up. This is a process known as contrastive learning that observes how things work in the world and projects different scenarios that it can apply. Robots can learn quickly through a smaller number of demos than was possible before.
The robot watches various YouTube videos of movement and actions such as a German Shepard playing with a ball.
Learning through practice in the real world means that the robot can adapt to changes that it could be confronted with in the future. When robots are trained solely through simulation, they are only prepared for that simulated environment.
Meanwhile, researchers at the Robotics Institute of Carnegie Mellon University are working on robots that will copy and assist humans, according to Technology Networks. The ‘In-the-Wild Human Imitating Robot Learning’ (WHIRL) project has an algorithm for “one shot visual imitation” that will enable it to perform everyday tasks.
The team used a standard robot model and added their software and a camera for observation, then it was soon able to learn more than 20 household tasks, such as opening and closing refrigerator doors and repositioning chairs. The robot needs to watch the task being performed only once, then it’s ready to practice until it gets it right.
Similar to the Dreamer algorithm robot, the WHIRL robot makes a break from simulation techniques to learning that uses an imitation model. The team claims it is scalable, flexible, and suitable for a home environment, where it can learn and perform a range of tasks.
Taking things to the next level, another new robotics project claims to have developed a robot that has learned to be self-aware. Science Daily reports that researchers at the Columbia University of Engineering and Applied Science have created a robot that can learn a kinematic model of itself without any human assistance.
The engineers started with a robotic arm surrounded by five live-feed cameras reporting back to the robot. This allowed the robot to move and learn its responses to motor commands, which stopped after 3 hours.
Self-learning of awareness in robots like this saves labor costs, as well as the ability to notice its condition and damage, compensate for damage, and call for assistance. This means that robots can function more efficiently in the real world and make better decisions.
As self-awareness is an animal quality, this could be a reminder of the recent news that a Google employee identified a Google AI as sentient. In the last week, the tech giant made the decision to terminate its contract with Blake Lamoine, according to CNN.
We are still a long way off the sinister robotic creations of Hollywood that we can find in everything from Blade Runner to Black Mirror, but that’s not to say we shouldn’t be concerned about particular robot news stories.
One such account can be found in The Guardian this week when a robot broke the finger of an unsuspecting child during a chess match in Moscow. This malfunction apparently occurred when the boy took his turn too quickly, as safety regulations state that players should wait for the robot to finish its move.
An accident like this has not happened before with the robot chessmaster, but parents will surely think twice before they allow their children to play with this robo-bully.
In other news
- Russia leaving the International Space Station. Roscosmos, the Russian space agency (and NASA’s counterpart), announced that it will leave the ISS after 2024. The agency plans to build its own space station called the Russian Orbital Space Station, or ROSS. “We will fulfill all our obligations to our partners, but the decision to leave this station after 2024 has been made,” Roscosmos chief Yuri Borisov said. However, Gizmodo points out that international sanctions against Russia for its invasion and war in Ukraine may make construction more difficult than it might otherwise. As explained in The Washington Post, the American and Russian sections rely upon one another, and with Russia’s departure, it will create a major void in the operation of the ISS. Russia runs the critical propulsion control systems. The announcement is a diplomatic disappointment, as it was just earlier in the month that NASA and Roscosmos had agreed to send each other’s astronauts/cosmonauts to the ISS.
- Scientists hack fly brains and dead spiders. As part of a research project devoted to restoring vision in people whose eyes don’t work, scientists have been experimenting on house flies. Interesting Engineering explains the research, which involves genetically modifying the flies and then injecting them with nanoparticles that could be controlled by an electromagnet. The goal of the research is to discover how to stimulate parts of the human brain associated with vision. Meanwhile, Science Alert describes how mechanical engineers at Rice University have discovered a way to get a dead wolf spider’s limbs to grip objects. Spiders use hydraulic pressure rather than muscles to move their legs, and in these experiments, the researchers used super glue and air to trigger this mechanism and manipulate the spiders’ limbs. This research could lead to the development of tiny, biodegradable robots in the future.
- Half of Earth’s minerals were formed by living things. In a major departure in our understanding of the planet’s geology, researchers have proposed a new taxonomic system for classifying minerals. As published in American Mineralogist and reported by Quanta Magazine, researchers Robert Hazen, Shaunna Morrison, and collaborators suggest that it’s important to classify minerals based on how they were formed rather than how they appear. They recognized that many minerals form exclusively from organic materials, while many others form in part through life processes, such as using oxygen produced in photosynthesis. While this research helps scientists understand earth processes and mineral formation, it may one day also help scientists with extraterrestrial geologic formations.
- Scammers targeting LinkedIn users with malware. A ring based in Vietnam is targeting LinkedIn users most likely to use Facebook advertising. According to Gizmodo, the scammers create fake messages encouraging users, often in digital media or marketing, to download malware that infects the users’ computers and seeks Facebook information. From there, the hackers use the login data to add themselves to users’ Facebook Business accounts so they can send out scammy ads under the cover of legitimacy. As quoted by Gizmodo, Meta spokesperson said they are “aware of these particular scammers, regularly enforce against them, and continue to update our systems to detect these attempts.”
Tip of the Week
Today, robotics is more accessible than ever. From DIY robot kits to educational coding apps, many tools are available to help you get started in this fascinating field. Whether you’re just dipping your toes in robotics or hoping to begin a career in the field, some essential tips will help you get up to speed as fast as possible.
Robotics is a field that requires forward thinking. While building a physical product can be incredibly exciting, it’s important to always keep your end goal in mind. How do you want your product to work? What do you want it to be able to do? How will you build it? Most importantly, how do you plan to finish it? Take the time to map out every single detail of your product, from its functionality to its aesthetic.
As you work on building your product, it’s essential to test your code and stay organized. Use a project management system to stay organized, so you don’t lose track of anything. Keeping everything organized and together will help prevent you from losing anything in the chaos.
Whether working with a team or another person, it’s important to learn how to collaborate and communicate effectively. Collaborating can help you build new skills, explore new ideas, and gain experience that you can build on for future robotics projects.