In recent years, with the rapid development of computer and network technology, the creation of many large-scale knowledge engines has greatly changed the human knowledge sharing and representation ecology. Inspired by this, some scholars at Cornell University and Stanford University have created a new, large-scale knowledge of robots that can be accessed freely by any device that is capable of performing tasks. Engine: Robotic brain (RoboBrain).
Robot brain: a large-scale knowledge engine for robots
In the human-oriented knowledge database, information search is a simple matter. We only need to input a few words on a computer or mobile terminal to get an answer. In many cases, fuzzy and default search can solve the problem. But for robots, things are not that simple. Even with the simplest tasks, robots require detailed and complete operational details. Based on the search results obtained by human-oriented search engines, robots are difficult to read. For example, in order to make the robot read the search results of "how to bake a cake", in addition to pointing out the conventional baking process and steps, it is necessary to provide a variety of detailed and relevant information, such as eggs can break; break the liquid flowing out of the eggs It must be contained in a vessel such as a cup; the cup can hold the liquid only when the cup is facing up; the water comes from the faucet, can be heated in a kettle or microwave oven, and can be mixed with egg liquid. Therefore, creating a large-scale knowledge engine for robots is not an easy task.
At a time when the prospects for the development of the artificial intelligence industry are generally optimistic, Cornell University and Stanford University have jointly created a robotic brain, a large-scale knowledge engine for robots, and created a website for this purpose. This is primarily a large-scale knowledge engine for robots to learn and share knowledge representations and perform multiple tasks.
The home page of the "Robot Brain" website shows that the knowledge engine learns various concepts mainly by searching the Internet and other data sources: "It can interpret natural language texts, images and videos; it can use its sensors to observe humans; They interact to learn things.†The creators believe that by learning and sharing large-scale knowledge, different robots can perform tasks more quickly and agilely in a variety of new situations. The creation of the "robot brain" changed the way the robot learns deeply. Traditionally, the focus of robot learning is rule learning. With the advent of the era of big data, robots began to move from rule learning to multimodal data learning.
Robot brain: making robot deep learning possible
The “robot brain†is essentially a large, crowd-sourcing database that stores a variety of sources of knowledge, including the physical interactions that robots engage in when completing tasks such as perception, planning, and control. , the knowledge base of the World Wide Web, and a variety of learnable knowledge representations created by leading robotics research groups.
The creation of the "robot brain" has largely changed the current state of big data mining over-reliance on single-text data modal sources. Subject to technical limitations, the data source for large-scale knowledge engines for humans is mainly "structured" textual modal data on the World Wide Web, and cannot effectively exploit other "unstructured" data modalities such as symbols, pictures, Video, audio, etc. “Structured data†mainly refers to data that can be logically characterized by the database, while “unstructured data†mainly refers to data that cannot be logically represented, mainly including multiple data modalities: text, pictures or images. , touch, XML, HTML, various charts or reports, audio and video information, and more.
According to statistics, in today's big data revolution, about 80% of the world's data is "unstructured data." Traditional computers were unable to process this huge amount of data before the advent of cloud computing. With the help of cloud-based information mining, storage, processing and presentation technologies, we can analyze and calculate them to some extent. As a result, the value of "unstructured data" has been greatly highlighted. Since the robot-oriented knowledge engine is largely composed of “unstructured dataâ€, which is essentially a source of data, how to effectively encode multiple different data modalities and build one that can A data representation-compatible knowledge representation framework is the biggest challenge facing the "robot brain." To this end, the "robot brain" resorts to the graphical framework to resolve conflicts and inconsistencies between different data modalities.
64V Battery Pack ,Lithium Battery Box,Lithium Power Pack,Jackery Battery Pack
Zhejiang Casnovo Materials Co., Ltd. , https://www.casnovonewenergy.com