22Astronauts

Igor Kulakov is the Co-Founder and CEO of MicroFactory (SimpleAutomation, Inc.), a San Francisco-based robotics startup building autonomous "factories-in-a-box" for precision electronic assembly.

Backing their bet that tabletop manufacturing cells beat humanoids for real-world reliability, MicroFactory is backed by Naval Ravikant and Hugging Face CEO Clément Delangue, using continuous human teleoperation feedback loops to reach near-100% precision.

MicroFactory: https://microfactory.com

Igor on X: https://x.com/ihorbeaver

Ilir on X: https://x.com/IlirAliu_

Ilir on LinkedIn: https://www.linkedin.com/in/ilir-aliu/

Timestamps:

0:00 Why we don't need humanoid robots for tabletop assembly

2:24 Inspired by "Back to the Future": Building pinball machines and telescopes at 15

6:30 Creating indie games and viral social media apps for 1M daily users

11:40 The pain of physical hardware: Running a wedding light startup in Ukraine for 9 years

21:16 Moving to San Francisco to be at the frontier of physical AI

32:06 Discovering the problem: Why hardware startups struggle with assembly

36:18 The viral photo frame demo that launched MicroFactory at Founders, Inc.

39:38 Why big AGI labs ignore precision and 99.9% reliability

43:07 Beyond pure data collection: Why reinforcement learning damages hardware

45:00 Human-in-the-loop DAgger: Correcting errors on Jetson Nano for $5 per retraining

52:15 Positive ROI vs. perpetual demos: Why shipping matters

62:41 Advice to founders: Change your environment if your experiments aren't working

Dr. Jiafei Duan is a Presidential Young Professor at the ⁨National University of Singapore (NUS), Director of the MAGIC Lab (Manipulation and General Intelligence Control), and Research Scientist at A*STAR.

Formerly a researcher at UW and the Allen Institute for AI (Ai2), Jiafei co-led the open-source "MolmoAct" and "MolmoAct2" robotics foundation models; action reasoning models designed to provide a transparent, fully reproducible alternative to proprietary "black box" physical AI systems.

MAGIC Lab: https://magic-ailab.github.io/

RoboPapers Podcast: https://www.youtube.com/@RoboPapers

Jiafei on X: https://x.com/DJiafei

Ilir on X: https://x.com/IlirAliu_

Ilir on LinkedIn: https://www.linkedin.com/in/ilir-aliu/

Timestamps:

0:00 Why true robotics progress requires radically open science

1:45 The RoboPapers podcast: Asking the hard research questions

3:34 Building my first mobile manipulator from PVC pipes at 16

7:19 The Leap Motion epiphany: Controlling plywood with Arduino

10:15 Writing the survey that introduced "Embodied AI" to Singapore

14:55 Why Singapore is investing heavily in physical intelligence

21:36 Moving from cognitive science to robotics at UW with Dieter Fox

24:51 AR2-D2: Collecting robot manipulation data using AR & inpainting

28:41 The PhD shift: Learning how to think and communicate science

32:49 Why robotics isn't ready for industry to consolidate yet

39:11 Vision researchers are migrating to ICRA: The field is shifting

41:26 Running open VLA models on a $200 SO-100 hobbyist robot

46:38 Leaving Ai2: The philosophy behind fully open data and weights

50:52 Launching the MAGIC Lab at NUS: Finding efficient model architectures

54:36 Why robotics won't have its ChatGPT moment without open source

My conversation with Prof. Dr. Majid Khadiv, Head of the ATARI LAB (Applied and Theoretical Aspects of Robot Intelligence) and Assistant Professor at TUM (the Technical University of Munich).

The ATARI Lab focuses on AI planning in dynamic environments, bridging optimal control with modern foundation models to make general-purpose humanoid robots safe, adaptable, and truly autonomous.

ATARI Lab: https://ce.cit.tum.de/en/aipd/

Majid on LinkedIn: https://www.linkedin.com/in/majid-khadiv-a92a2a62/

Ilir on X: https://x.com/IlirAliu_

Ilir on LinkedIn: https://www.linkedin.com/in/ilir-aliu/

Timestamps:

0:00 Building national humanoids in Iran without resources

2:11 Why playing soccer beats being a tech kid

3:03 How my sister proved you can break through the entrance exam

5:10 Thinking deeply about a single math problem for five hours

7:02 Visualizing physics: Why Newtonian mechanics just clicked

8:38 Why traditional schooling fails curious minds

10:50 Should AI tutors replace the classroom experience?

18:48 Discovering the intersection of control and dynamics in robotics

20:30 Building the Surena III humanoid from scratch with PhD students

23:44 Honda's Asimo vs our 98kg robot: The engineering reality

26:01 How the 2015 DARPA Robotics Challenge humbled everyone

28:55 Why grades don't reflect true deep-thinking ability

34:01 How Ludovic Righetti transformed my research mindset at Max Planck

37:09 The price to pay for staying at the forefront of academia

52:19 Why general purpose humanoids are the holy grail

58:26 The hand is the pinnacle of human intelligence

62:03 Fusing language models with evolutionary search for physics planning

64:27 Why safety and OOD cases will keep humanoids out of homes for years

70:19 Advice to my younger self: Have confidence in your curiosity

Andrew Kang is the Co-Founder and CEO of ⁨RoboStrategy⁩ and Managing Partner of Mechanism Capital.

RoboStrategy already holds concentrated positions in industry-defining companies including ⁨Figure AI, ⁨Apptronik, ⁨Standard Bots⁩, and Dyna Robotics.

It is the first publicly listed vehicle purpose-built to give investors liquid exposure to private robotics and physical AI giants, applying the MicroStrategy capital-markets playbook to automation.

RoboStrategy: https://www.robostrategy.ai

Andrew on X: https://x.com/Rewkang

Andrew on LinkedIn: https://www.linkedin.com/in/rewkang/

Ilir on X: https://x.com/IlirAliu_

Ilir on LinkedIn: https://www.linkedin.com/in/ilir-aliu/

Timestamps

0:00 Why we built a public fund for private robotics

0:35 Building the MicroStrategy for robotics

1:22 Escaping the tiger parent trap

2:11 Video games are elite training for game theory and critical thinking

6:30 Learning arbitrage in RuneScape and Neopets

7:59 Don't believe experts just because of their titles

10:24 Why investing is just a video game with real money

12:40 Harvard nutrition professors don't know what they're talking about

18:48 Recreating global finance from first principles

19:57 Why robotics is the early days of crypto all over again

21:50 Realizing humanoids are no longer sci-fi

22:58 Why I wrote a $19M check when every VC told me not to

27:24 How SPVs are democratizing venture investing

30:55 Bet on the founder and their execution speed, not the pitch deck

36:39 The math behind a $50 trillion physical labor market

41:44 Applying the MicroStrategy playbook to private markets

45:48 Humanoids are the GPUs of physical labor

47:48 Building a new SoftBank for the robotics era

50:51 Why robotics startups need engineers for attention

54:35 Don't be satisfied with surface level answers

Eliot Horowitz is the Founder and CEO of Viam and Co-Founder of MongoDB.

Viam is building the universal software platform to bridge the gap between AI and physical hardware, running everything from automated boat-sanding systems to AI sonar fleet operations.

Viam: https://www.viam.com

Eliot on X: https://x.com/elliothorowitz

Ilir on X: https://x.com/IlirAliu_

Ilir on LinkedIn: https://www.linkedin.com/in/ilir-aliu/

Timestamps:0:00 The early days of 20 megabyte hard drives1:18 Writing video games on a Radio Shack computer at age five6:24 Rockets, airplanes, and why software is the hardest part of hardware9:03 The power of iteration velocity11:35 Becoming an entrepreneur by accident: MongoDB15:00 Product vs. utility: Why MongoDB won19:10 Stepping down and tinkering in the COVID era22:41 Why I bought a $25k industrial robot arm for my New York apartment28:28 Taking a big swing: Building a universal platform34:16 From an idea to 1,000 production robots41:00 Why trying to build both hardware and software is a trap43:53 Real world deployments: Tooling for operators vs technologists47:02 The power of dogfooding coffee and salad robots in the office49:16 How LLMs will actually code the future of robotics52:14 It’s not about the product, it’s about the outcome53:36 Why nobody spends enough time on education

Evan Beard is the co-founder and CEO of Standard Bots, building AI-trained industrial robot arms designed to automate real factory work, not demos:In this episode, Evan shares a founder path that started in software startups and Y Combinator long before robotics. Seeing engineers at YC building giant robots convinced him this was a “cheat code on life” and he taught himself mechanical and electrical engineering to make it possible 

We talk about the early years spent building prototypes in a small apartment, running more than one hundred calls with manufacturers, and discovering the same problem every time. Companies wanted automation, but robots were too expensive and too difficult to program 

Evan explains why Standard Bots focuses on usable automation instead of chasing hype. The goal was not a slightly better robot but something ten times easier to deploy, trained through demonstration and physical AI rather than complex programming 

We also discuss surviving the near-death phase, raising funding weeks before running out of money, rebuilding electronics during the chip shortage, and learning electrical engineering while debugging exploding circuit boards remotely 

A conversation about persistence, customer obsession, and why the future of robotics will be decided by real ROI on factory floors, not humanoid spectacle.

Grace Brown is the founder and CEO of Andromeda, building social companion robots designed for aged care and healthcare environments:

In this episode, Grace shares a founder journey that started long before a company existed. She had been building robots since her teenage years, but the real motivation appeared during COVID lockdowns. People in care facilities were physically supported, yet emotionally isolated. The goal became simple. Build a robot people actually want around them.

We talk about learning outside the curriculum, reaching out to mentors early, and running long real-world pilots before the product was ready. Instead of waiting for perfection, Andromeda deployed unfinished systems into nursing homes to understand behavior, trust, and human reaction.

Grace explains why the hardest problem in social robotics is not intelligence but comfort. A robot can function technically and still fail if people feel uneasy around it. Design, personality, and interaction determine adoption more than raw capability.

We also discuss building a company as a young founder, hiring commercial leadership early, and how moving into the US startup ecosystem changed the pace of decisions and iteration.

You don't want to miss this one!

Camilla Mazzoleni is the co-founder and Chief Product Officer of FORGIS, a Zurich-based startup building an AI operating layer for industrial automation:

In this episode, Camilla shares a founder journey shaped long before robotics.

She grew up as a competitive alpine skier, leaving home early to train at elite level. Discipline, pressure, and repetition defined her daily life. Losing was normal. Learning how to reset and keep going became second nature.

After an injury ended her professional sports career, that same intensity moved into engineering. Camilla talks about discovering robotics through building, not theory. Working on factory floors, programming robots across vendors, and seeing firsthand how slow and fragmented industrial automation really is.

We talk about how Forgis came to life at ETH Zurich. Why hardware is not the bottleneck in factories. Why software fragmentation is. And how Forgis sits on top of existing systems as a hardware-agnostic, edge-based AI layer that upgrades how factories operate instead of tearing them apart.

A great conversation about discipline, switching paths, and why Europe’s manufacturing future depends on intelligence, not replacement.

Stephen James is the founder and CEO of Neuracore, and Assistant Professor of Robot Learning at Imperial College London:

In this episode, Stephen shares his path from growing up in Wales to spending a decade at Imperial, a postdoc at Berkeley, and eventually founding Neuracore. Not because he wanted to be a startup founder, but because he kept running into the same problem again and again: every robotics team rebuilding the same infrastructure from scratch.

We talk about what actually slows robotics teams down, why data pipelines matter more than clever algorithms, and how Neuracore aims to become the infrastructure layer that lets teams focus on deployment instead of plumbing.

Stephen also reflects on imposter syndrome, work ethic, moving between academia and industry, and why Europe can and should build its own robotics infrastructure instead of copying Silicon Valley playbooks.

A very honest conversation about building foundations, not hype, and why scaling robotics is mostly about removing friction.

Steve Xie is the founder and CEO of Lightwheel AI, building the simulation and synthetic data layer powering the next generation of embodied AI and humanoid robotics.

In this episode, Steve shares a rare founder journey that starts far from robotics. From studying physics at Peking University, struggling to stand out, and rebuilding confidence through sheer consistency, to a PhD at Columbia and an early failed startup built out of love for his dog. A detour that taught him the cost of building without a business model.

We talk about his path through Cruise, NVIDIA, and NIO, where he led large-scale simulation efforts for autonomous driving. Steve explains how those years shaped his conviction that simulation, data quality, and evaluation are the real bottlenecks in physical AI.

He then breaks down how Lightwheel AI came to life. Why sim-ready assets matter more than solvers. How synthetic data actually closes the sim-to-real gap. And why robotics teams hit a ceiling without proper evaluation and scaling infrastructure.

A deep conversation about resilience, delayed gratification, and why the hardest part of building is often unlearning what made you successful before.