iia/0930 Robotics panel

• Robotics at Scale Shipping autonomous systems that work outside the lab — the engineering, regulatory, and commercial hurdles of deploying robots and drones at scale.

  • meta

    • [Panel]
    • [Sustaining Intelligence]
    • 9:30 AM
    • Panelists
      • Person/Navid Aghasadeghi moderator
        • AI & Robotics
        • Founder & CEO
      • Person/Alexander Wissner-Gross
        • Founder & Investor
        • last person to get 3 degrees, math, physics, CS, then phd from harvard, crazy
      • Person/Annika Thomas Robot/ics/Space
        • MIT Dept. Mechanical Engineering
        • PhD Candidate
      • Person/Hasan Khan
        • Formlabs making Print/er/3d which is a form of Robot/ics
          • Product & AI
  • topics
    • 09:30 what's different this time? was there one experience or demo or product you saw which made you think, this time it's actually going to work?
      • Person/Alexander D. Wissner-Gross
        • computer vision got solved
        • willow robotics guys

          • asked them: what is the hardest? single number one problem
        • action models are on scaling law curve
        • AI/Model/Action are on Law/Scale/Curve
        • AI/Model/World are on Law/Scale/Curve
        • The Singularity is not one thing, it started with AlexNet and the curves are smooth from there
        • we're basically "there" for some definition of "there" at this moment
        • don't think AI/The Bitter Lesson
      • Person/Annika Thomas
        • Vision/Computer is solved in a sense, but we are working with limited compute IRL

          • we have multiple world robots
          • there are lots of limitations in the hardware in order to make these
          • embedded systems are harder
      • Person/Hasan Khan
        • Google/Glass -> Apple/Vision/Pro progression is amazing
      • Person/Annika Thomas
        • SLAM Acronym = simultaneous localization and mapping

          • they have to solve the estimation problem of where we are at in reference to the chair
          • she thinks slam is solved in small-scale indoor settings, but outdoors with dynamic settings
      • Person/Alexander D. Wissner-Gross
        • he just wrote Person/Alexander D. Wissner-Gross/Book/Solve Everything

          • he's talking about the idea that all that there is is to pour on more scale
          • "do you think asymptotic leaps are necessary?"
      • Person/Annika Thomas
        • difference between AI/LLMs and AI/Robot/ics is Data/Set/Availab/ility
        • she talks about a paper that came out years ago where everyone came together to combine data, but nobody uses it because it's too messy
  • Person/Navid Aghasadeghi let's go deeper into data. how do you think about how Robot/ics can deal with separate layers of intelligence stack?
    • Person/Alexander D. Wissner-Gross
      • elephant in room: what do you pre-train off of?
      • we are seeing pre-training off of YouTube and pre-recorded data
        • you can get really far on this
        • first-person, third-person doesn't make much of a difference
      • then you fine-tune on custom data
      • he thinks there's no data scarcity - frontier labs have better access to the compute to make use of the corpora that are already available
      • we are post-data scarcity from my standpoint Take/Hot
  • Person/Navid Aghasadeghi ok, so intelligence is about to be solved, but we don't see millions of robots around us. it's really hard to deploy robots at scale. Person/Hasan Khan has deployed lots of 3d Printers
    • Person/Hasan Khan
      • 3d Printers are autonomous goal-oriented systems
        • they are facing similar challenges to Robot/ics
        • people have Hesitation about AI because they tried, old, bad AI
      • a lot of people are worried about Job/Loss or Skill/Atrophy
  • Person/Annika Thomas
    • we've always had one part that works on perception, localization, navigation, and now we have people working on AI/Model/End/to/End, and she's worried about that in terms of AI/Safe/ty and Repro/duc/ibility
      • when we put Swarms of Robots out there, I want to know why a Mistake happened
  • 09:50 Person/Alexander D. Wissner-Gross
    • the implied trade-off between safety and capability is a false dichotomy
      • we've seen that in Tesla/FSD with more recent iterations, they introduced modularity
      • it's very much possible to maintain modularity and Super/vision and and maintain End/to/End "Differentiable" solutions
  • 09:52 Person/Navid Aghasadeghi is safety a "feature" of the system Alex?
    • Person/Alexander D. Wissner-Gross
      • you get there by having rigorous AI Evals
    • 09:53 Person/Annika Thomas
      • with Diff/eren/tia/bil/ity we will get there
      • I want to emphasize that you made a great point about AI Evals. Many of those systems in the past were based on Average Performance, but we need Cert/if/iab/ility, and Case/Edge matters.
  • 09:54 Person/Navid Aghasadeghi we think about replacing physical jobs with Robot/ics, but where are the opportunities to expand the market, such as for applications in Space?
    • Person/Annika Thomas
      • this is what got me into robitcs - space
      • NASA administrator jared something said we will focus on settlements on the Moon
        • to do that we need Robot/ics/Space and autonomy
        • they just patented radiation shielding with bricks that are fully constructed on the moon autonomously Cool
        • AI/Response
          • You’re likely thinking of Person/Jared Isaacman, not the NASA Administrator.
          • NASA Administrator is Person/Bill Nelson
          • Isaacman has been publicly discussing long-term human presence beyond Earth, including the Moon, in the context of private spaceflight and collaboration with NASA.

            
            For actual NASA policy language about Moon settlements, the authoritative framing comes from the Artemis Program, which explicitly states goals like:
            

          • “sustainable human presence on the Moon”
          • building infrastructure that enables longer-term habitation

            
            Primary sources you can use:
            

          • https://www.nasa.gov/artemis/
          • https://www.nasa.gov/humans-in-space/artemis/

            
            If you want the exact quote you heard, it’s probably from:
            

          • a recent panel/interview featuring Isaacman discussing future human expansion
          • or a paraphrase of NASA’s “sustainable presence” language rather than a literal “settlements” quote from Nelson

            
            Key distinction:
            

          • NASA (Nelson): “sustainable presence”
          • Isaacman / private sector: more willing to say “settlements” explicitly
  • 09:59 Person/Alexander D. Wissner-Gross
    • he says "the economy is cooked"
      • even doctors
      • 'belmals cost disease'
    • next weekend - ProRL will focus first professional Robot/ics Marathon.
      • this is the first time in US history where we will have robots in the streets in the US/MA/Boston/Seaport Cool
        • Robot/Olympics
      • "come to the combine 2026-04-19 Sun"
  • 10:04 comment from crowd - I just saw an OpenClaw bot to Blender designing 3d/Models, how will that impact things?
    • Person/Hasan Khan
      • when Robot can Hill Climb, that's the new point.
  • 10:05 we talked about solved vs unsolved problems. What are unsolved problems in LLMs vs need for AI/Model/World Question
    • Person/Annika Thomas
      • LLMs are the bridge between the environment and the Robot/ics between perception and Plan/ing.
      • something like LIDAR has uncertainty associated with it, world models help with that?