Andrew Barry on Why Dexterity Is the Next Breakthrough in Physical AI

Andrew Barry on Why Dexterity Is the Next Breakthrough in Physical AI

Physical AI is moving quickly. But Andrew Barry says one of the biggest unlocks in robotics is not just getting robots to move through the world. It is getting them to touch, grasp, adjust, and manipulate the world with real dexterity. In this episode of Automated, Brian Heater speaks with Andrew Barry, co-founder and CTO of Generalist, about how the company is building general intelligence for the physical world and why dexterous robots may be the starting point for far more capable automation. Andrew explains why Generalist is focused on the tasks that are both difficult and valuable. Robots have made major progress in mobility, but their ability to manipulate objects is still limited. If robots can solve dexterity, they can become useful in a much wider range of real-world environments. The conversation explores how Generalist is collecting massive amounts of real-world manipulation data. Andrew describes the handheld data capture devices the company built, why they chose that approach over teleoperation, and how thousands of devices have helped them scale a much richer data set for robot learning. Brian and Andrew also discuss the commercial side of physical AI. Andrew explains why the company is not just chasing impressive demos, but benchmarking against real tasks people are already paying for today. That distinction matters because a viral robot demo is not the same thing as a deployable robotic system. They also dig into one of the most surprising parts of modern robot learning: improvisation. Andrew shares the moment when a robot picked up a baggie with the opposite hand from the one it had been trained on, completed the task anyway, and left the team realizing something very different was happening inside the model. The episode also covers Generalist’s GEN-1 model, the parallels between robotics and the early GPT era, why flexible objects like cables are so difficult to automate, what data flywheels may actually look like in robotics, and why robots sometimes learn human mistakes from the data they are trained on. Finally, Andrew reflects on his path from Boston Dynamics to the Broad Institute and then to Generalist, explaining how work in molecular biology, machine learning, transformers, and robotics all shaped the way he thinks about building intelligence for the physical world. Connect with Andrew Barry https://www.linkedin.com/in/andy-barry [https://www.linkedin.com/in/andy-barry?utm_source=chatgpt.com] Learn more about Generalist https://generalistai.com/ [https://generalistai.com/?utm_source=chatgpt.com] We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org [podcast@automate.org] You can find the transcript and more episodes of Automated at automated.fm [http://automated.fm]. Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify. https://www.youtube.com/@automatedpodcast [https://www.youtube.com/@automatedpodcast] https://podcasts.apple.com/us/podcast/automated-with-brian-heater/id1837762221 [https://podcasts.apple.com/us/podcast/automated-with-brian-heater/id1837762221] https://open.spotify.com/show/60olq6brlBEIJWggx2fMR6 [https://open.spotify.com/show/60olq6brlBEIJWggx2fMR6] You can also find us on: LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ [https://www.linkedin.com/showcase/automated-podcast-by-a3/] Instagram https://www.instagram.com/automatedpod/ [https://www.instagram.com/automatedpod/] Subscribe to the Automated Newsletter: https://www.automate.org/automation/automated-newsletter [https://www.automate.org/automation/automated-newsletter] ---------------------------------------- Hosted on Acast. See acast.com/privacy [https://acast.com/privacy] for more information.

Daniel Rausch on How Alexa Was Rebuilt for the AI Era

Alexa is entering a very different era. For years, voice assistants were built around rules, scripted responses, and carefully designed commands. But with the rise of large language models and generative AI, Amazon had to rethink what Alexa could be and how people might use it. In this episode of Automated, Brian Heater speaks with Daniel Rausch, Amazon’s Vice President of Alexa and Echo, about Alexa+, the company’s AI-powered evolution of its voice assistant. Daniel explains why the shift from traditional voice assistance to foundational AI assistance required a full rearchitecture of the technology behind Alexa. The conversation explores how Alexa moved from a deterministic system to one powered by more than 70 models, why customers do not care which model is working behind the scenes, and how Amazon thinks about choosing the right AI tool for the job. Brian and Daniel also discuss one of the biggest questions around AI assistants: trust. Daniel explains why Alexa is designed to understand that it is AI, why it should help people prioritize human relationships, and why guardrails matter as assistants become more conversational, personal, and ambient in the home. They also get into the smart home, where Daniel says Alexa+ is changing how people interact with connected devices. Instead of needing to know the right command or app, people can speak naturally, whether they are unlocking a door, checking a Ring camera, controlling lights, or asking for help while cooking. The conversation also covers Echo hardware, privacy controls, personality styles, language and dialect differences, AI’s impact on robotics, and why Daniel sees Amazon as an invention machine at a moment when AI is moving faster than ever. Connect with Daniel Rausch https://www.linkedin.com/in/danielrausch [https://www.linkedin.com/in/danielrausch?utm_source=chatgpt.com] Learn more about Alexa+ https://www.amazon.com/alexaplus/dp/B0CXRRF584 [https://www.amazon.com/alexaplus/dp/B0CXRRF584?utm_source=chatgpt.com] Learn more about Amazon Echo devices https://www.amazon.com/b?ie=UTF8&node=210779651011 [https://www.amazon.com/b?ie=UTF8&node=210779651011&utm_source=chatgpt.com] We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org [podcast@automate.org] You can find the transcript and more episodes of Automated at automated.fm [automated.fm] Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify. You can also find us on: LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ [https://www.linkedin.com/showcase/automated-podcast-by-a3/?utm_source=chatgpt.com] Instagram https://www.instagram.com/automatedpod/ [https://www.instagram.com/automatedpod/] ---------------------------------------- Hosted on Acast. See acast.com/privacy [https://acast.com/privacy] for more information.

Daniela Rus on Humanoid Robots, Physical AI, and the Future of Robotics

Physical AI is moving fast. But Daniela Rus says the future of robotics will not be defined by viral humanoid robot demos alone. The real challenge is building robots that can understand the physical world, make safe decisions in real time, and work reliably outside controlled lab environments. In this episode of Automated, Brian Heater speaks with Daniela Rus, Director of MIT CSAIL, about humanoid robots, self-driving cars, embodied AI, on-device AI, robot learning, and why the next wave of artificial intelligence needs to move beyond the cloud and into the physical world. Daniela explains why humanoid robots are exciting, but not ready for prime time. A robot may look impressive in a short demo, but operating safely and consistently around people requires common sense, physical understanding, and real-world adaptability that robots still do not fully have. The conversation also explores why self-driving cars remain one of the hardest problems in robotics. Daniela breaks down the long tail of autonomous driving, from bad weather and unpredictable human behavior to the messy edge cases that make real-world deployment so difficult. Brian and Daniela also discuss why the future of AI robotics may depend on smaller, more efficient AI models that can run directly on devices. If a car is moving at 60 miles an hour, it cannot wait for the cloud to decide what to do next. For robotics, speed, safety, energy use, and reliability all point toward a hybrid future where AI runs both in the cloud and on the machine itself. Daniela also shares why physical AI needs more than video data. Robots interact with the world through forces, torques, motion, contact, and uncertainty. For many tasks, robot learning requires a deeper understanding of physics, not just visual imitation. The episode also moves into some of the most fascinating frontiers of AI and robotics, including Daniela’s work with Project CETI and the effort to better understand sperm whale communication using machine learning, robotics, and large-scale data collection. Finally, Daniela talks about AI systems that could help design robots from natural language prompts, why engineering constraints can drive creativity, what octopus intelligence can teach us about decentralized robots, and why this moment in robotics feels like the future researchers imagined decades ago is finally arriving. Connect with Daniela Rus https://www.csail.mit.edu/person/daniela-rus [https://www.csail.mit.edu/person/daniela-rus?utm_source=chatgpt.com] Learn more about MIT CSAIL https://www.csail.mit.edu/ [https://www.csail.mit.edu/] Learn more about Liquid AI https://www.liquid.ai/team/daniela-l-rus [https://www.liquid.ai/team/daniela-l-rus?utm_source=chatgpt.com] Learn more about Project CETI https://www.projectceti.org/ [https://www.projectceti.org/?utm_source=chatgpt.com] We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org [podcast@automate.org]. You can find the transcript and more episodes of Automated at automated.fm [http://automated.fm]. Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify. Subscribe to the Automated Newsletter: https://www.automate.org/automation/automated-newsletter [https://www.automate.org/automation/automated-newsletter] You can also find us on: LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ [https://www.linkedin.com/showcase/automated-podcast-by-a3/?utm_source=chatgpt.com] Instagram https://www.instagram.com/automatedpod/ [https://www.instagram.com/automatedpod/] ---------------------------------------- Hosted on Acast. See acast.com/privacy [https://acast.com/privacy] for more information.

Matthew Johnson-Roberson on Why Physical AI Still Has a Missing Piece

Physical AI is moving fast. But Matthew Johnson-Roberson says robotics is still missing something fundamental. The field has data, models, and momentum, but it still does not have the simple learning objective that helped language models scale so quickly. In this episode of Automated, Brian Heater speaks with Matthew Johnson-Roberson, founding dean of Vanderbilt’s College of Connected Computing, about why physical AI may not follow the same playbook as large language models. Matthew explains why robotics still feels stuck between promise and deployment. We still do not live in a world where you can look out your window and see robots everywhere. That gap is not just about hype. It is about the difficulty of building systems that can learn from physical experience in a way that actually scales. Brian and Matthew also discuss what self-driving taught the broader automation world, why last-mile delivery still has not cracked scale, and what Amazon’s long arc with Kiva robots reveals about how real hardware progress actually happens. The conversation also explores healthcare, where Matthew says AI scribes are already making a real impact, even as outdated infrastructure like fax-based record sharing shows how much friction remains. That experience also helped inspire Patients.app, the startup he co-founded after watching how much clinician time gets lost to documentation. They also get into the tension between startups and academia. Matthew argues that startups are powerful vehicles for scaling known solutions, but much worse fits for decade-long research questions that still do not have clear answers. Finally, Matthew reflects on building Vanderbilt’s new College of Connected Computing, why higher ed can take on 30- and 40-year problems in a way few other institutions can, and how AI agents have changed his own workflow so dramatically that he says he has not directly written a line of code in three months. Connect with Matthew Johnson-Roberson https://www.linkedin.com/in/mattkjr [https://www.linkedin.com/in/mattkjr?utm_source=chatgpt.com] Learn more about Vanderbilt’s College of Connected Computing https://computing.vanderbilt.edu/bio/matthew-johnson-roberson/ [https://computing.vanderbilt.edu/bio/matthew-johnson-roberson/?utm_source=chatgpt.com] Learn more about Patients.app https://patients.app/ [https://patients.app/?utm_source=chatgpt.com] We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org [podcast@automate.org]. You can find the transcript and more episodes of Automated at automated.fm [http://automated.fm]. Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify. Subscribe to the Automated Newsletter: https://www.automate.org/automation/automated-newsletter [https://www.automate.org/automation/automated-newsletter] You can also find us on: LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ [https://www.linkedin.com/showcase/automated-podcast-by-a3/] Instagram https://www.instagram.com/automatedpod/ [https://www.instagram.com/automatedpod/] ---------------------------------------- Hosted on Acast. See acast.com/privacy [https://acast.com/privacy] for more information.

Sergey Levine on Why Real-World Data Will Define Physical AI

Physical AI looks closer than ever. But the hardest part in robotics is not getting a machine to do one impressive task on camera. It is building systems that can improve from real-world experience, handle edge cases, and scale across different robots and environments. In this episode of Automated, Brian Heater speaks with Sergey Levine of Physical Intelligence about why robotics has reached an inflection point, and why progress now requires more than great models in a lab. Sergey explains why the next phase of robotics will depend on something much less flashy than a viral demo: collecting the right real-world data, learning from it efficiently, and building systems that improve through deployment. The conversation explores what makes a robot experience useful in the first place. Sergey describes a concept borrowed from child psychology called the “zone of proximal development,” where the best learning happens when a system is challenged just beyond what it can already do. For robots, that means creating environments where they can succeed, fail, adapt, and improve. Brian and Sergey also discuss how the bottleneck in robotics is changing. Basic motor skills are improving fast. The harder problem now is judgment. A robot may be able to clean dishes, but if it drops a clean plate on the floor, it still has to understand that the plate needs to be washed again. That kind of common sense remains one of the biggest unsolved challenges in physical AI. They also dig into one of the biggest debates in robotics right now: data. Sergey argues that real-world data collection is not the impossible obstacle many researchers once assumed. In fact, he believes the long-term path to better robots is more practical than people think. Deploy systems, collect experience, improve the model, and repeat. The conversation also covers why Physical Intelligence is focused on a general intelligence layer rather than a single-narrow product, why robots should not just be treated as metal versions of people, and what surprised Sergey most about controlling very different robot platforms with the same model. Finally, Sergey reflects on why Physical Intelligence is structured more like a lab than a traditional startup, why experimentation matters so much in modern AI, and how we may one day look back on this era as the moment AI moved beyond internet data and into the physical world. Connect with Sergey Levine https://www.linkedin.com/in/sergey-levine-5a31a24 [https://www.linkedin.com/in/sergey-levine-5a31a24?utm_source=chatgpt.com] Learn more about Physical Intelligence https://www.physicalintelligence.company/ [https://www.physicalintelligence.company/?utm_source=chatgpt.com] We’d love to hear from you. Have thoughts or guest suggestions? Reach us at podcast@automate.org [podcast@automate.org]. You can find the transcript and more episodes of Automated at automated.fm [automated.fm] Unlock full access to Automated and explore everything automation. Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify. Subscribe to the Automated Newsletter: https://www.automate.org/automation/automated-newsletter [https://www.automate.org/automation/automated-newsletter?utm_source=chatgpt.com] You can also find us on: LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/ [https://www.linkedin.com/showcase/automated-podcast-by-a3/] Instagram https://www.instagram.com/automatedpod/ [https://www.instagram.com/automatedpod/] ---------------------------------------- Hosted on Acast. See acast.com/privacy [https://acast.com/privacy] for more information.