A division of the Toyota Research Institute (TRI) has unveiled new robotics capabilities to solve complex problems in the home. In particular, TRI robotics have been able to train robots to understand and act in complex situations that confuse most other robots, including recognizing and responding to transparent and reflective surfaces in various circumstances.
The TRI demo video shows that the developed system allows robots to generalize in different scenarios, including in different houses. The video was released on National Selfie Day and ironically plays this holiday – you can see how the robot takes pictures of itself as it demonstrates its skills.
“Our goal is to create robotic capabilities that enhance, rather than replace, human capabilities,” said Max Bajracharya, vice president of robotics at TRI. “Teaching robots how to work at home poses particular challenges because of the diversity and complexity of our homes, where small tasks can become big.”
While a person can easily distinguish between an object and its reflection, transparent or reflective objects commonly found in the home, these same tasks confuse modern robots. Since most robots are programmed to respond to objects and geometries in front of them without regard to the context of the situation, it is easy to fool them with a glass table, a shiny toaster, or a transparent cup.
“To overcome this, TRI robotics have developed a new teaching method that allows us to perceive the three-dimensional geometry of the scene, as well as detect objects and surfaces,” continued Bajracharya. “This combination allows researchers to use large amounts of synthetic data to train the system.”
The use of synthetic data also reduces the need for time-consuming, expensive, or impractical data collection and labeling processes.
While no system is perfect, this announcement expands the knowledge base to help robots move and work reliably at home ultimately. This technical advance allows the robot to quickly learn from “programmable data” – synthetic data that can be recreated. Also, robots will be able to learn from past failures. Thus, the progress made by TRI is a promising milestone for the unit itself and robotics around the world.
A source: Toyota