Which specific practice or focus has been most important for achieving reliable, scalable robotics deployments, and why?

The answers emphasize reliability, safety, and real-world validation as crucial for scalable robotics, though they differ on whether design, data, accuracy, or operational focus is most important.

10 answers

Answers

  • Marc Lore

    Founder & CEO · Wonder

    Ease of maintenance

    Marc Lore says the biggest aha was designing robotics for production that never goes down, easy to clean and easy to maintain, since kitchen staff rather than robotics experts must maintain it.

    Inc. · Watch on YouTube
  • Mark Papermaster

    CTO and EVP · AMD

    Full stack optimization

    Mark Papermaster says full stack optimization is needed to achieve the latency and reflex response required in physical AI, since you can't get that without it.

    Imagination in Action · Watch on YouTube
  • Pete Florence

    Co-Founder & CEO · Generalist

    Focus on goals

    Pete Florence says being really focused on the goals they are trying to do, and being less precious about the methods that get them there, helps conjure simplicity.

    The Generalist · The Generalist · Listen at 32:56
  • Michelle Lu

    Co-Founder · Vsim

    Accuracy

    Michelle Lu says accuracy is the key limitation to address, because the simulation must match reality as closely as possible so that policies perform better when deployed to real robots.

    Robot Talk · Robot Talk · Listen at 8:12
  • Aaron Edsinger

    CEO & Cofounder · Hello Robot Inc

    Low-force, low-inertia joints

    Aaron Edsinger says the first layer of robot safety is physical design, noting Stretch's joints are lower force and lower inertia, so there are no pinch points and it won't topple like a bipedal humanoid.

    Robot Talk · Robot Talk · Listen at 17:55
  • Xavier (Tianhao) Chi

    Co-Founder · Mbodi Ai

    Field validation

    Xavier Chi says a lot of their testing came down to validating what's needed in the field, visiting factories and warehouses to confirm their technology could solve manual processes, which he frames as de-risking the investment through deployment.

    TechCrunch · Watch on YouTube
  • James Kuffner

    Chief Technology Officer · Symbotic

    Redundancy

    James Kuffner says his team builds redundancy into the system so failures fail gracefully and safely, citing redundancy in software, hardware and communications as what makes the robots safe.

    The answer starts at 37:10.

    Soft Robotics Podcast · Listen on Apple Podcasts
  • Chelsea Finn

    Co-Founder · Physical Intelligence

    Large real-world data

    Chelsea Finn says there is no getting around it: robots need training on large amounts of real-world data to perform in real-world situations, and she notes even lower-quality data helps because diversity teaches the model more ways to complete a task.

    AI Podcast Summaries from Transcripted.ai (VIDEO) · AI Podcast Summaries from Transcripted.ai (VIDEO) · Listen at 1:58
  • Reid Hoffman

    Co-Founder, Executive Board Chair · Manas Ai

    Redundant safety precautions

    Reid Hoffman says redundant safety precautions let you build a safe robot system even when parts of the software, like the neural network, are unpredictable or unreliable.

    Reid Hoffman · Watch on YouTube
  • Joe Lonsdale

    General Partner · 8vc

    High reliability

    Joe Lonsdale says the other important frontier is mastery-level performance, reaching high numbers of nines of reliability such as 99.99% so models can be deployed at scale in production.

    Joe Lonsdale · Watch on YouTube

More Robotics questions

View all →