Video So get this, Boston Dynamics, the non-evil robot-maker, went ahead and taught one of its “Spot” robo-dogs how to talk using ChatGPT. Now, for those who don’t know, ChatGPT is one of those massive language models that can predict the next bit of text based on what it’s given. It’s like one of those autocomplete algorithms on steroids. And it turns out, Boston Dynamics was curious to see how they could use this technology to make Spot make decisions in real time. Pretty cool, right?
They decided to test it out by turning Spot into a robot tour guide. This little dogbot could roam around, check out objects in its environment, describe them using a Visual Question Answering (VQA) model, and even answer questions from the tour audience. It was like having a little improv actor in the form of a robot. They provided a script, and the LLM filled in the rest on the fly. Talk about advanced technology!
But of course, they didn’t stop there. They wanted Spot to look like it was having a conversation with the audience, so they added a speaker and a microphone. They even gave it a gripping tool that moved like a puppet’s mouth. To enhance the illusion, they dressed up the gripper with silly costumes and stuck googly eyes on it. I mean, why not, right? Gotta make it entertaining!
Now, if you’re curious to see how well this talking robo-dog actually works, check out the image below. That’s Spot, in all its talking glory. And if that’s not enough for you, there’s even a video of Spot trying to interact with humans. Trust me, you’re gonna want to watch it!
Boston Dynamics talking robodog tour guide – Click to enlarge
Now, as amazing as all this sounds, the Boston Dynamics team did encounter some interesting moments during the process. For instance, when they asked the robot who its founder, Marc Raibert, was, the robot responded with, “I don’t know. Let’s go to the IT help desk and ask!” And guess what? It actually went to the IT help desk. I mean, talk about independent thinking!
But that’s not all. They also asked Spot to identify its parents, and it pointed to the “old Spots” in their office, telling them that those were its “elders.” Creepy yet kinda cute, I must say.
What’s even more impressive is how well the language model stayed in character, even when they gave it absurd personalities. The team found that “snarky” or “sarcastic” personalities worked really well for the robot. They even got it to go on a “bigfoot hunt” around the office, asking random people if they had seen any cryptids. I mean, that’s just plain hilarious!
Now, it’s important to note that ChatGPT does have its flaws. Sometimes it gives incorrect information, like when they asked about BD’s “Stretch” logistics bot, and it responded that it’s for yoga. And there were times when the conversation got a bit awkward due to delays between questions and answers. Plus, if OpenAI gets overwhelmed or the internet connection goes down, well, the robot might not be as entertaining. But hey, nobody’s perfect, right?
Despite these minor glitches, the Boston Dynamics team is pretty excited about the results. They believe that being able to talk to robots and have them understand and act on our commands will revolutionize how we interact with them. From being tools and guides to companions and entertainers, robots could become an integrated part of our lives. And honestly, that’s a world I wouldn’t mind living in.
So, get ready folks, because a future filled with talking robots might just be closer than we think. And when that day comes, we’ll be able to have some pretty interesting and hilarious conversations with these mechanical wonders. I can’t wait!