Robot Talk · 3 Nov 2023 · 27m · Episode 60
Carl Strathearn, a research fellow at Edinburgh Napier University, discusses the design and application of realistic humanoid robots, specifically Euclid and Bodie, for healthcare and social care. He explains the engineering challenges of creating lifelike features, such as Euclid's lip-syncing capabilities and Bodie's gelatin-based eyes with artificial muscles, while emphasizing the importance of narrow, task-specific applications over general-purpose AI. Strathearn highlights the rigorous regulatory approval processes required for testing these robots with vulnerable populations and critiques the media's misunderstanding of current AI capabilities.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulation
A machine learning model was trained on a dataset of similar-sounding words to predict lip positions from acoustic data for speech synthesis.
“And it was really about focus on the sound wave. Hmm. And it's what it has to do is train a model on certain words, so it will look like a machine learning library, basically, a data set of similar sounding words.”
Speaker 2 · Hear it at 12:07
Engineering & simulationBert
The team used the Bert model for its superior ability to understand context in task-based interactions, rather than using larger foundation models.
“But in our project, we use Bert, because it's much better at understanding context, which you need in kind of task-based interaction.”
Speaker 2 · Hear it at 11:15
Engineering & simulationgenerative model
A generative model was used to create a dataset of common sense knowledge from recipes, allowing the robot to answer questions about missing ingredients.
“So we had like a group of interns who we imagined these recipes singing them like communicating with the robot for the 1st time. … we used. We created this data set off kind of the back of the original recipes that you can generate from. Okay. So use like a generative model to, so now you can kind of ask questions and get a response from this common sense knowledge database.”
Speaker 2 · Hear it at 1:55
early-stage companies
Other shows on the same ground, closest first.
Brian Heater, Managing Editor at A3 and host of the 'Automated' podcast, discusses the nuances of covering robotics and AI, emphasizing the importance of long-form interviews to uncover genuine insights from industry pioneers like Mark Roper and Rodney Brooks. He highlights the critical gap between research-scale robotics and commercial viability, noting that while Amazon has achieved massive scale, most humanoid startups are still far from reliable, safe production. Heater expresses optimism for the future of automation driven by human-centric applications, such as exoskeletons for workers, prosthetics inspired by personal history, and devices for Parkinson's patients, rather than just general-purpose humanoid robots.
Emeritus Prof. Rodney Brooks, CTO of Robust AI and founder of iRobot and Rethink Robotics, discusses the necessity of physical intelligence in robotics, arguing that reactive systems and force control are superior to pure position control or simulation. He critiques the current hype around humanoid robots and large language models, emphasizing that successful products like the Roomba and Baxter industrial robot relied on pragmatic engineering and series elastic actuators rather than complex AI. The episode concludes with a discussion on the societal implications of automation and the enduring value of fundamental hardware principles.
Dr. Jan Liphardt, Stanford bioengineering professor and co-founder of Open Mind, discusses the transition of robotics from sci-fi to practical application through his company's modular AI architecture, OM1. He explains how Open Mind’s approach uses discrete, Lego-like functional blocks rather than monolithic models to enable transparency, debugging, and personalized social intelligence in robots like Iris. Liphardt highlights the rapid maturation of hardware components, such as reliable robot hands and self-charging capabilities, predicting widespread household deployment by late 2026, while emphasizing safety, transparency, and the importance of human-robot interaction in education and companionship.
Hosts Daniel and Forbode discuss a research project from Columbia University that aims to overcome the 'uncanny valley' in humanoid robotics by creating a robot named Emo with a highly expressive, human-like face. They detail the hardware design, which uses 26 actuators and soft silicon skin to mimic the 30 muscles of the human face, and the software training process where the robot learned expressions by watching itself in a mirror and 'binge-watching' YouTube videos of human faces. The conversation explores the concept of 'co-expression' in human interaction, the ethical implications of creating robots that are too convincing, and the potential integration of large language models like ChatGPT to provide conversational context for the robot's reactions.
Dr. Kate Darling, research lead for ethics and society at the AI Institute, discusses the intersection of robotics and social science with hosts Steve Crow and Mike Oitzman. They cover the recent RoboBusiness 2023 event, highlighting the industry's focus on humanoids like Agility Robotics' Digit and Figure AI's robot, as well as the integration of Generative AI in robotics. Darling shares her perspective on human-robot interaction, using the Stop & Shop Marty robot and public perception of autonomous vehicles like Waymo and Cruise to illustrate the social challenges of deploying robots in shared spaces.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Robot Talk. The episode itself belongs to its publisher.