army of More than 4,000 Dogg Marches robots It is a mysterious menacing spectacle, even in simulation. But it may point the way for machines to learn new tricks.
The virtual robot army is developed by researchers from ETH Zurich In Switzerland the chip maker nvidia. Use roaming robots to train algorithm which were then used to control the legs of a real robot.
In simulation, machines are called Anyone—Encounter challenges such as cliffs, steps, and steep slopes in a virtual scene. Each time the robot learned to beat a challenge, the researchers presented a harder challenge, pushing the control algorithm to be more complex.
From a distance, the resulting scenes resemble an army of ants writhing across a large area. During training, the robots were able to master going up and down stairs with sufficient ease; More complex obstacles took longer. The slopes have proven to be very challenging, although some virtual robots have learned how to slide on them.
When the resulting algorithm was ported to a real version of ANYmal, a four-legged robot roughly the size of a large dog with sensors on its head and a detachable robot arm, it was able to navigate stairs and blocks but struggled with problems at higher speeds. Researchers blamed inaccuracies in how sensors perceive the real world compared to simulations,
Similar types of machine learning can help machines learn all kinds of useful things, from Packet sorting to me Sewing clothes And crop harvest. The project also reflects the importance of simulation and dedicated computer chips for future application progress Artificial intelligence.
“At a high level, a really fast simulation is really nice,” he says. Peter AppelHe is a professor at the University of California at Berkeley and one of the founders a variable, a company that uses artificial intelligence and simulation to train robotic arms to pick and sort things for logistics companies. He says the Swiss and Nvidia researchers have “got some really nice velocity speeds.”
AI has shown the promise of training bots to do realistic tasks that can’t be easily written into software, or that require some kind of adaptation. Being able to make sense of awkward, slippery, or unfamiliar things, for example, isn’t something that can be written into lines of code.
4000 imitation robots have been trained using Learning EnhancementAn AI method inspired by research on how animals learn through positive and negative feedback. As the robots move their legs, an algorithm determines how this affects their ability to walk, and adjusts control algorithms accordingly.