Every year, Boston Dynamics It strives to take its developments to another level, and there is ample evidence of this. On this occasion, the company has taken advantage of ChatGPT so that spothis famous robot dog, be able to talk to humans.
The end result is quite peculiar, and even a little creepy. In a video shared on YouTubethe company belonging to Hyundai showed how it turned Spot into a kind of guide that interacts with people and talks with them while showing them the facilities around them.
To accomplish these tasks, Boston Dynamics not only experimented with ChatGPT and other AI-based tools, but made physical modifications to the robot dog. Specifically, it incorporated a microphone and speaker so that it can receive questions or comments from the humans around it and respond to them. To talk to the device, the command “Hey, Spot” was implemented.
But that is not all. Boston Dynamics engineers developed a method for Spot’s gripper to function as a “mouth.” Thanks to its ability to detect and follow moving objects, the arm located on top of the robot always points towards the person who is speaking to it; while the clamp moves to accompany the voice. To complete the outfit and make it a little more comical, they added plastic eyes and a small hat.
Spot, Boston Dynamics’ robot dog, takes advantage of ChatGPT and generative AI
One of the most interesting points of this Boston Dynamics proof of concept is that the robot dog can be used with different personalities. At first we see him using nuances and the tone of voice of a British butler, but then he adopts expressions that emulate those of a teenager or a “Shakespearean time traveler”, among other types of identities or temperaments.
In this experiment, Boston Dynamics used the ChatGPT API—first with GPT-3.5 and then with GPT-4—as well as other massive open source language models. Spot was also provided with image recognition software and speech-to-text and text-to-speech converters. Likewise, experts took advantage of what is known as “technology engineering.” prompts” to get the robot dog to talk.
“Inspired by a Microsoft approach, we asked ChatGPT to pretend to type the following line in a Python script. We provide English documentation to the LLM in the form of comments. We then evaluate the output as if it were Python code. The LLM has access to our autonomy SDK, a tour site map with one-line descriptions of each location, and the ability to say phrases or ask questions.”
Boston Dynamics, on how they made their robot dog “talk” using ChatGPT.
A striking fact is that Boston Dynamics found a positive use for ChatGPT’s problematic hallucinations. That is, the ability of AI to invent on the fly when it does not know the real answer to a question asked. “In this case, we did not need the tour to include precise facts, but rather to be entertaining, interactive and full of nuances,” said the robotics experts.
On its official blog, the company explained with more technical details how he got his robot dog to use ChatGPT to communicate with humans. Of course, this is just a first step in a project that, in the not too distant future, could expand drastically.
“We are excited to continue exploring the intersection of artificial intelligence and robotics. These two technologies are a great combination. A world where robots They can understand what you say and turn it into useful actions “It’s probably not that far away,” insists Boston Dynamics. For now, the firm has not mentioned whether it plans to experiment with something similar with Atlas, its humanoid robot.