Spot the Robot Dog: Talking with Human-like Fluency

Home Forums Product Design Spot the Robot Dog: Talking with Human-like Fluency

  • This topic is empty.
  • Creator
  • #3106

      In the world of robotics, Boston Dynamics’ Spot has already dazzled enthusiasts with its wide range of skills, from conducting sniff tests to cutting a rug on the dance floor. This versatile robot dog can even oversee construction sites, ensuring the utmost safety. But the latest revelation from Boston Dynamics will make dog lovers’ hearts skip a beat – Spot can now talk, and it does so with flair, courtesy of ChatGPT.

      Spot the Robot Dog: Talking with Human-like Fluency

      In a recent showcase, Boston Dynamics showcased Spot’s newfound linguistic talents, which include multiple accents and personalities, such as British and American dialects. When you hear Spot converse, it’s as if a real human is speaking. The robot dog’s lifelike, viper-like mouth opens and closes, emitting voices that are far from the typical robotic monotone. But what’s the intended purpose for this talking robot dog? Boston Dynamics envisions it as a tour and museum guide, ready to regale visitors with historical narratives and anecdotes about artworks and objects.

      Spot the Robot Dog: Talking with Human-like Fluency

      The Magic Behind Spot’s New Skill

      Boston Dynamics’ endeavor to turn Spot into a captivating conversationalist relies on the prowess of language models like ChatGPT. These models excel at predicting and completing text, transforming them into exceptional tools for generating human-like speech. Boston Dynamics was inspired by this text prediction capability, as well as the advent of Visual Question Answering (VQA) models, which can caption images and answer queries about them. Through a blend of hardware integrations and software models, Spot was transformed into an articulate tour guide, capable of engaging in coherent conversations that resemble those of real humans.

      To empower Spot, Boston Dynamics equipped it with a camera, speaker, and microphone. These enable Spot to perceive the visitors, comprehend their questions and remarks, and reply audibly. To enhance the conversational experience, OpenAI’s ChatGPT, initially with gpt-3.5 and later upgraded to gpt-4, was integrated into Spot’s system. The team, consisting of Matt Klingensmith, Michael McDonald, Radhika Agrawal, Chris Allum, and Rosalind Shinkle, divulges that to make Spot’s responses audible, they employed a text-to-speech tool.

      Spot the Robot Dog: Talking with Human-like Fluency

      Spot’s Impressive Tour Guide Skills

      As evidenced in the video demonstration, Spot is adept at guiding visitors through an exhibit or a piece of artwork by furnishing them with pertinent facts and data about the subject. During the demo, Boston Dynamics decided to test Spot’s knowledge by asking about Marc Raibert, to which the robot dog powered by ChatGPT responded, “I don’t know. Let’s go to the IT help desk and ask!” Spot then autonomously made its way to the IT help desk to inquire about Marc Raibert. Boston Dynamics clarifies that they didn’t program Spot to seek assistance, indicating that this behavior emerged naturally.

      Spot the Robot Dog: Talking with Human-like Fluency

      Another intriguing instance occurred when they asked Spot about its parents. In response, Spot embarked on a journey to the office where Boston Dynamics’ earlier iterations of Spot, known as Spot V1 and Big Dog, were stationed. It then informed the inquirer that these older robot dogs were its forebears.

      The Boston Dynamics team was pleasantly surprised by how well the language model stayed in character, even when assigned absurd personalities. They discovered that snarky or sarcastic personas worked remarkably well, and they even sent the robot on a whimsical bigfoot hunt around the office, asking unsuspecting passersby if they had spotted any cryptids.

      Boston Dynamics plans to further explore the capabilities of Spot the robot dog and its ChatGPT language model, including tasks that can be assigned to the robot through spoken commands. The future of robotics is looking increasingly fascinating and engaging with these advancements.

    • You must be logged in to reply to this topic.