Spot, a robot dog created by Boston Dynamics, can now ‘speak’. Thanks, ChatGPT

Here is how the Hyundai-owned US robotics firm trained Spot to answer questions.

Spot, a robot dog tour guide created by Boston Dynamics, can now ‘speak,’ as the Hyundai-owned US robotics design firm leveraged OpenAI's ChatGPT to ‘train’ the robot to answer questions, and generate responses about the company's facilities.

How does Spot ‘speak’?

Boston Dynamics' 'Spot' robot (Image courtesy: Boston Dynamics)

As per The Verge, to make Spot ‘talk,’ Boston Dynamics used ChatGPT along with some open-source large language models (LLMs) to carefully train the robot's responses. The engineers then outfitted it with a speaker, and added text-to-speech capabilities.

The team gave Spot a ‘very brief’ script; by combining the script with the imagery it gets from the cameras on its gripper and body, it gathers more information about what is in front of it, and generate a response accordingly.

As per Boston Dynamics, Spot essentially captures images and answers questions about them, doing so using Visual Question Answering models.

A demonstration…

In reality, Spot actually ‘mimics’ the act of speaking. As seen in a demonstration uploaded by Boston Dynamics on YouTube, the ‘tour dog’ opens its ‘mouth’ to answer questions; it is actually the text-to-speech capabilities that generate responses, while the speaker blares out the answers.