Author(s): Raghu Gopal
Researchers from the Massachusetts Institute of Technology (MIT) have developed a robot that plays Jenga using technology that could be used to assemble consumer products or separate recycling.
Advances in artificial intelligence have been picking up pace, giving machines the ability to learn and adapt, changing their behaviour over time. But in the three-dimensional physical world, machines still aren't as dexterous as people.
Researcher Nima Fazeli and his colleagues at MIT have created a robot that has a basic understanding of physics and a usable sense of touch. The team was looking to explore how robots could learn about their environment by using more than just machine vision.
To do this, the engineers turned to the game of Jenga. They evaluated the game's hierarchical learning approach to acquire manipulation skills for the robot, and compared them with other methods. Jenga is a simple game that involves removing wooden blocks from a precariously assembled tower without causing it to topple over. The researchers explained that "Jenga is a quintessential example of a contact-rich task where we need to integrate with the tower to learn and to infer block mechanics and multimodal behavior by combining touch and sight".
The team used the robot arm, an Intel RealSense D415 camera and an ATI Gamma force-and-torque sensor mounted "at the wrist" for the experiment. Machine learning software was then created for the learning approach. By combining vision, touch, and a model of real-world physics, the robot can learn to play Jenga more efficiently than would be possible otherwise. In testing, the approach outperformed conventional machine-learning methods.
The drawbacks of robots are well understood and were highlighted recently when a hotel built in 2015 in Japan, known for its worker robots, had to "fire" over half of the nearly 250 machines, as they weren't delivering the benefits that had been expected. The management team decided it would be cheaper to replace them with humans.
But the trajectory is clear: robots are progressing at an astonishing speed and MIT's clever robot gives us a glimpse into the future robot world that's evolving.