A leaked internal document revealed that the retail giant Amazon is secretly developing an upgraded version of its household robot Astro so that it can better understand what it observes and act more intelligently on things. React.
The project, internally code-named “Burnham,” adds a layer of “smart and conversational voice interface” to Astro, the documents show. Amazon described the new technology, called contextual understanding, as “our latest and most advanced artificial intelligence technology designed to make robots smarter, more useful, and better at conversations,” the filing said.
In documents related to Burnham, Amazon describes an Astro product that costs $995. A “Burnham Plus” with standard home monitoring features will cost $24.99 a month, while a Burnham Plus with an Amazon Ring doorbell camera will cost $34.99 a month, according to the document.
Burnham is quite versatile and can support a range of other products. The technology “can remember what it sees and understands,” and draw deeper meaning from those interactions, the document says. It can also have a conversation about what it sees and understands, and take appropriate action based on that, using large language model (LLM)-based techniques common in apps like ChatGPT.
For example, if Burnham sees a fire left on or a tap running unattended, it will find the owner and bring it to their attention. If an elderly person slipped and fell in the kitchen, Burnham could help check to make sure he was okay and call others to help. If it’s an emergency, Burnham will automatically call the appropriate emergency number for assistance.
The owner can ask Burnham where he left the key. It can also check to see if the kitchen window was open last night and monitor if the kids have friends over after school.
An internal document reads: “Simply put, our robots already have a strong body, and the next step is a brain. A robot with Burnham can understand all kinds of trivial things that happen in the home like a human being, without having to worry about everything. Everything is encoded because ‘common sense’ knowledge is implicit in the data on which the language model is based.”
The move represents the next stage of growth for Astro. Astro, Amazon’s heavily invested home-monitoring robot powered by its Alexa smart assistant, has so far failed to live up to Amazon’s lofty expectations. Despite years of investment and the work of hundreds of people, Astro’s capabilities remain mediocre and difficult to buy because it’s still invitation-only, even 18 months after launch.
It is also the latest example of Amazon integrating generative artificial intelligence (AIGC) and LLM techniques into its existing products and services, as rivals such as Microsoft and Google lead the way in the AI chatbot market. Amazon is reportedly planning to upgrade its Alexa voice technology to have ChatGPT-like features, while it also wants to build artificial intelligence tools to automatically generate pictures and videos for advertisers.
Amazon Chief Financial Officer Brian Olsavsky said last month that more investment is moving from its core logistics business to such AI and LLM technologies.
In an email, an Amazon spokesman said Astro “is off to a promising start” and that requests for invitations “remain strong,” though he did not provide numbers. “We’re excited by the potential of AI technologies and look forward to using them to create new experiences for Astro that will delight customers and make their lives easier in the future,” the spokesperson added.
Amazon’s development of Burnham was inspired by a large research paper in the field of language models. As LLMs grow in size, they begin to demonstrate “new skills in reasoning and problem-solving,” Amazon said in the filing. Inference occurs when an AI model makes predictions based on new data.
Amazon, for example, has found that with LLM, household robots can go beyond simple tasks like finding out if a light is left on or a door is open. Instead, they can now handle more complex tasks like detecting broken glass on the floor, knowing that shards of glass are dangerous and need to be cleaned up before someone steps on it.
Amazon conducted a concept demonstration to test the ability of its new home robot to demonstrate reasoning and problem-solving. The test was successful. “You can imagine our excitement as we learn what this means for our robots. We know we have a long way to go before Burnham is integrated into our products,” the company’s internal document reads. go, but we’ve learned enough to believe it will work.”
Like other teams at Amazon, the Astro unit has faced significant cost pressures over the past year, according to people familiar with the matter. The team scrapped internal plans to release a low-cost model last year and instead announced a new focus on small business security.
The Future of the Burnham Project
Burnham may not be limited to just one product. In an internal document, Amazon defined it as a set of core technologies, rather than a product itself, and said it expected Burnham to “appear in a range of future robots.”
Burnham may test the limits of large language models as Amazon attempts to move its AI technology from the digital realm into real-life physical environments, such as into home hubs.
According to the filing, Amazon appears to believe that Burnham is a dramatically improved version of a home monitoring robot that “gives families peace of mind that everything is fine at home.”
“Burnham’s friendly personality, social awareness and fluent conversational skills make him highly available and would be a welcome addition to any family,” the document added.
