Earlier this year I wrote about the latest output from MIT’s Computer Science and AI Laboratory (CSAIL). They developed a robot that was capable of collaborating effectively with other robots, and tested out their creations in a bar environment, with robots taking on the role of waiters and barman.
They were able not only to coordinate orders from customers but also ensure they took the best route to and from customers to ensure the bar ran smoothly.
A similarly collaborative approach was showcased by the CARLOS team, who developed a robot capable of operating in the semi-autonomous environs of a shipyard.
Central to the process was the ability for the robot to ‘learn’ on the fly rather than rely purely on pre-programed instructions. Researchers at the University of Maryland are working on just such a learning robot.
“We call it a ‘robot training academy,’” the researchers say. “We ask an expert to show the robot a task, and let the robot figure out most parts of sequences of things it needs to do, and then fine-tune things to make it work.”
The team also used a hospitality style environment to showcase what their robot was capable of. At a recent conference, the robot, which was made in partnership with Rethink Robotics, first observed a human making a cocktail, before then copying them and mixing the drink themselves.
How to train a machine
The learning process involves the robot being able to associate video footage of humans performing certain deeds, with the corresponding actions they themselves are then required to perform to replicate them.
In early tests, the robots were able to perform simple maneuvers after watching a few thousand YouTube videos. Whilst this may sound time consuming, it’s actually less so than manually training the robot.
The next stage is to partner with some larger manufacturing companies to adapt the technology for factory use. The initial feedback suggests that these companies are looking for ways to speed up the learning process for their existing machines.
“At many companies it normally takes a month and a half or more to reprogram a robot,” the researchers say. “So what are the current AI capabilities we can use to shorten this span, maybe even in half?”
The project is further evidence of the growing ability of robots to both learning independently and also work in close proximity to humans. Whilst the robots still struggle with spoken or written instructions, this is something the researchers are confident will soon be overcome.