News

Toyota has Developed a New Generation of Home Nanny Robots, Which Can Perform 85% of Complex Human Tasks

Toyota Research Institute (hereinafter referred to as tri) has successfully developed the latest generation of Home Nanny Robots: busboy, which can perform 85% of complex human tasks! The combination of cloud level learning ability and cloud level learning ability can make robots grow exponentially.
Home Nanny Robots can not only clean the floor and furniture, but also get back the drinks you need from the refrigerator. The robot itself will continuously perceive the surrounding environment through sensors, predict the safe path, and then formulate the walking or running route according to this understanding.

Home Nanny Robots


At Toyota’s press conference, “bus boy” demonstrated his abilities, such as being able to pick up glasses from a bright reflective table top and put them in the sink, accurately pick up glasses from a highly reflective marble table top, and clean tables and floors.

Home Nanny Robot


Every family is a special complex environment. The content of housework is too diverse and cumbersome. It is difficult to do a standardized customized program to meet all family needs. Therefore, in our human opinion, the “smallest” housework needs the support of the most advanced AI and machine learning technology for robots.

Electric Claws


For example, we humans can simply recognize a transparent glass and distinguish a tea cup from its reflection projection, but it is a technical problem for robots. According to the instruction document of Toyota Research Institute (TRI), “the most common transparent or reflective objects at home are very confusing to robots”, “because most robots are programmed to respond to the objects and geometry in front of them, regardless of other background conditions, they are easy to be fooled by glass tables, shiny toasters or transparent cups”.

Electric Grippers


Home Nanny Robots,The Toyota “bus boy” uses more advanced AI and machine learning technology to “perceive the 3D geometry of the scene and detect objects and surfaces”. This combination enables researchers to use a large amount of synthetic data to train the system. So this is a robot that needs continuous “training” and upgrading. For robots, being able to recognize and pick up a transparent glass or distinguish a tea cup from its shadow is a more complex task than playing the violin or basketball.


Toyota Research Institute officially released the introduction video of robot learning system:

https://youtu.be/ISTdLXq0TeY


In order to solve the diversity faced by Home Nanny Robots in the home environment, tri teaches robots to perform arbitrary tasks with various objects (such as grasping objects, opening and closing doors, etc.) through humans, rather than programming robots to perform specific predefined tasks with specific objects. In this way, the robot learns to associate what it sees with what it learns. When the robot sees a specific object or scene again, it knows what action it can take on what it sees, even if the scene changes slightly.

Electric hands
Take things from refrigerator after learning


Researchers can use the robot’s sensors in front of the computer screen to see the content viewed by the robot in real time in 3D. Researchers can also choose different behaviors to guide the robot, and then annotate it to associate various parts of the scene with the behavior, such as teaching the robot how to hold the handle and open the refrigerator door.

robot's sensors
Opening refrigerator door


For robots, it is very difficult to operate and navigate in the home environment. Because each house is unique, it has different combinations of different configuration objects, and these combinations may change over time. The robot itself will continuously perceive the surrounding environment through sensors, predict the safe path, and then formulate the walking or running route according to this understanding.

Home Nanny Robots
Walk after predicting the route


The robot learning system only needs to understand the surrounding things related to the executing behavior, and can teach the robot to associate the behavior with any scene, object and voice command without inputting the object model or map in advance. At present, tri’s system can successfully perform about 85% of relatively complex human level tasks, and each task consists of about 45 independent behaviors.

Home Nanny Robots
Learning through voice commands


The next development focus of Toyota Research Institute is “robots for household chores”, which will trigger robot(Home Nanny Robots) developers all over the world to follow into this field, which is beneficial to overcome the “difficulties” in more household chores. Benign competition can promote the development speed, accelerate the technological progress and lower the price. It is hoped that more smart and capable “bus boys” will be on the market to help the elderly do housework.

 

Prev:

Next:

Leave a message