The robot dog with the ChatGPT ‘brain’ and Google ‘voice’
Coming to a construction site near you, a new generation of robots from Boston Dynamics is taking things to the next level and opening up implications for the future of work
First, they built a robot. Then they gave it a brain. Or, first they built an AI, then they gave it a body. Spot, Boston Dynamic’s robot dog, has been overhauled to integrate a ChatGPT brain.
The project was carried out in partnership with Levatas, a company that specialises in the automation of visual inspection processes. Spot was originally built to carry out simple reconnaissance missions, collecting huge datasets of observations. This might include regular checks of equipment at an industrial installation, surveying a building site, or security scans. Foster + Partners used it for regular site surveys during its work on the Battersea Power Station redevelopment in London.
Source: Foster + Partners
However, a key limitation has been that Spot requires a technically knowledgeable handler as there’s no simple way to provide instruction or query the data that is gathered. Spot can look at something and capture information about it. But it doesn’t – on its own – have the ability to understand what it’s looking at or make independent recommendations about it.
Natural language commands
In response, Boston Dynamics and Levatas have been working to give Spot a ChatGPT ‘brain’ and the ability to respond using Google Assistant’s voice technology.
Credit: Boston Dynamics/Levaras
This integration means that anyone can now talk to Spot using natural language commands to direct, query and obtain previously hidden insights. For example, when Spot is asked a question about its tasks, ChatGPT can interpret it and formulate a response by analysing the data it has collected. It also adds a new layer of safety – rather than a single remote operator, any human will be able to quickly pause or alter the robot’s autonomous mission by simply asking it to do something else.
The next step is to work how to use GPT-4 to help Spot solve complex tasks, not just follow commands. These types of large language models are currently limited by the computing power required – many industrial facilities don’t have internet access throughout the site.
No technical training needed
Eventually, the team believe that these complex brains will be able to run within the robot itself, removing the need for a link to the cloud. This means that the next generation Spot will be able to, based on a core mission, work out how best to perform it without any technical training at all.
Spot isn’t the first example of ChatGPT moving beyond text-based responses to be given a human-like voice. D-ID, an Israeli start-up, launched a beta version of a web app that allows users to converse with ChatGPT through a customisable avatar.
Source: D-ID
Similarly, the video collaboration platform Bubbles enables users to talk to ChatGPT through a human simulation. AI-assisted work is here, and with the rise of video conferencing avatars as a way to combat Zoom fatigue, it might not be too long before you don’t have to show up to meetings at all.
Oh – and don’t watch this before bed.
Read Imogen Privett’s complete design round-up in our Innovation Zone here.
The Innovation Zone is exclusively for Global Partners, Community and Corporate Members of WORKTECH Academy. Click here to join.
