Google robot autonomous dialogue

In early November, Google, for the first time, released an internally developed software tool that reduces the amount of work required to train robots to perform new tasks.

The tool is called “Code as Policy,” or CaP. It is available on GitHub under an open source license.

Before a manufacturer can deploy a robotic arm in a factory, it must customize the system to its own requirements. This process usually involves writing custom code. For example, a manufacturer’s developer might write code that instructs the robot arm to pick up products from the production line and place them in a box.

The company must write custom code each time it wishes to train a robot to perform a new task. In addition, adapting software written for one robot to run on other machines often requires manual modifications. As a result, the development process can involve a lot of time and effort.

Google says its newly unveiled CaP tool can save developers time by automatically generating robot configuration code. caP uses artificial intelligence models to generate code. In recent years, Google and other companies have developed advanced artificial intelligence systems that can write software based on user prompts. Using such artificial intelligence systems, CaP can generate code that enables robots to perform user-specified tasks.

Google researchers tested CaP’s capabilities in a series of internal evaluations. In one test, researchers evaluated whether CaP could teach a robot how to change the position of toy blocks on a table. After receiving a command to “arrange the blocks in a square in the middle,” CaP successfully generated code that enabled the robot to rearrange the blocks.

According to Google, the tool could also be used to train the robot for a variety of other tasks.” In a blog post, Google research intern Jacky Liang and research scientist Andy Zeng explained, “CaP allows a single system to perform a variety of complex and diverse robotic tasks without the need for task-specific training. “

The artificial intelligence system that powers CaP was not originally designed to generate robot configuration code. According to Google, its researchers used a method known as “odds learning” to train these systems to perform the tasks.

Teaching an AI system to perform a new task usually involves providing it with a large number of examples of how the task should be performed. By learning a few times, researchers can train AI systems with just a few examples, thereby speeding up development. Google researchers trained CaP by providing it with examples of how to translate natural language commands into robot configuration code.

CaP writes its software in the Python programming language. In addition to generating new code, the tool can automate common tasks using software libraries, which are pre-packaged sets of code. Google says its approach has proven more effective than existing methods of configuring robots to perform new tasks.

“Our experiments demonstrate that outputting code leads to improvements in generalization and task performance compared to directly learning robot tasks and outputting natural language actions.” Liang and Zeng elaborated, saying.

In addition to CaP’s code, Google has released a benchmarking tool to support further research. The benchmarking tool will allow researchers to more easily compare how well different AI systems perform robot-related tasks.

About Google Inc. Area 120 Startup Incubator Enters Enterprise Shop
Google is creating a division called “Area 120,” which is headed directly by company executives Don Harrison and Bradley Horowitz, to support employees in their internal startups. Employees can submit BP applications to join, and after being approved, their self-organized teams can work full-time for new ideas, and if successful, they can set up separate companies and get Google investment. There are comments that Google’s move is to prevent employees from jumping ship startups.

Google set up a new department Area120 to support employees’ internal entrepreneurship

The Google internal business incubator called “Area 120” by company executives Don Harris (Don Harrison) and Bradley Horowitz (Bradley Horowitz) in charge. At a recent all-hands meeting, the two executives introduced the new organization.

“Area 120 is located in Google’s new San Francisco office building, and Google executives hope that Area 120 will allow entrepreneurial employees to stay with the company longer, while also identifying big ideas. The incubator is primarily for employees and aims to maintain a startup atmosphere and connect with some hot entrepreneurs. The details of the incubator’s operation have not yet been determined, but the general framework is currently as follows: Google teams can apply to join the incubator, work full-time for a few months, and submit a concrete business plan; after that, they have the opportunity to receive a letter of intent from Google to build a new company, with Google as an investor.

Google has a well-known tradition of encouraging employees to spend 20% of their working time on their favorite projects, and the new incubator is named “Area 120” as a tribute to this system. In theory, Area 120 allows employees to work full-time on their favorite projects.

Google has already set up two companies, Google Ventures and Google Capital, which also aim to provide financial support for startups and keep some entrepreneurial former Google employees within the company. It is not clear whether these fund companies will cooperate with Area 120, according to speculation that Area 120’s funding should come from outside Google’s corporate development budget.