Scaling Robotics

In Episode 10, we speak with Matthew Lee, who has spent the last decade running field operations and sustainment at three autonomy companies: Aurora (on-road autonomous trucks), Phantom Auto (teleoperated forklifts), and most recently Scythe Robotics (autonomous commercial lawnmowers, deployed across 20+ states). Before robotics, he spent two decades in automotive repair, working his way up from technician to service manager.I believe that field engineering is the most criminally underrated function in running and scaling a fleet of robots. Matthew has spent his whole career on that side of the problem, and has watched companies live or die (like Phantom Auto) based on how seriously they took it.We get into why sustainment is the function most robotics startups underinvest in, and how Phantom Auto's failure to iterate on hardware in the field helped sink the company. Why field ops works on a totally different timescale than engineering. And how to build a real support stack from tier one all the way up to your developers, including who you should hire for each tier and what data you need to surface when a customer calls in.There are also a few tidbits on how to run your software deployments and canary releases. Enjoy this one, it's a top to bottom analysis of how to run field ops from someone who's experienced the struggle first hand.𝐋𝐈𝐍𝐊𝐒- Miru Blog (including episode transcripts): https://www.mirurobotics.com/blog- Spotify: https://open.spotify.com/show/1xr38bFEhk7MSf8UvmPUBo?si=761aeb03b4464fd1&nd=1&dlsi=fa48d636a0f74df6- Apple Podcasts: https://podcasts.apple.com/us/podcast/scaling-robotics/id1857724488𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒(00:00:04) – Introduction(00:01:09) – Overview of Autonomy Use Cases(00:03:50) – What Is Field Ops in a Robotics Company(00:05:52) – Why Sustainment Is Critical(00:07:59) – Building a Support System from First Call to Resolution(00:14:05) – Tier Two Support and Escalation(00:20:11) – The Tier Two Agent Profile(00:24:52) – Feeding Field Data Back to Product and Engineering(00:30:49) – Canary Releases and Software Rollouts(00:34:22) – Networking Challenges in Field Deployments(00:38:41) – The John Henry Problem: Operator Adoption(00:42:49) – Advice for Scaling Your First Fleet(00:45:23) – Closing and Where to Find Matthew

This EpisodeIn Episode 8, we speak with Cem Ersoz, Director of Robotics Software at Simbe Robotics. Simbe builds Tally, an autonomous retail robot that scans millions of shelves per week, giving retailers real-time visibility into product availability, out-of-stocks, and shelf accuracy. With thousands of Tallys deployed across major retailers in more than 10 countries, operating daily alongside untrained store employees across grocery, club, farm supply, and home improvement, Simbe runs one of the largest commercial Physical AI fleets in the world.We discuss what actually changes when your fleet crosses key thresholds. At a certain scale, you can no longer SSH into a robot to fix it, and that forces a complete rethink of how you build. We get into the difference between 95% autonomous and truly autonomous, why improving 9s of reliability is so hard, and how Simbe ships software to thousands of robots they can never directly touch.Plus, why A/B testing a robot fleet isn't like optimizing a SaaS funnel. There are things that no dashboards can capture, such as whether people actually feel comfortable around a robot moving faster through the store. We also get into what a truly representative canary rollout looks like at this scale, and what Cem learned when hardware he had trusted for years finally failed on robot number 2,000.Timestamps(0:04) Introduction (1:32) Fleet Scale and Deployment (3:57) Lessons from Operating Alongside Humans (6:56) Building Software for Thousands of Robots (10:55) Testing, Deployment, and Reliability (13:58) A/B Testing in Robotics vs. SaaS (17:06) Supporting Legacy Hardware (22:20) AI Coding Tools in Robotics (25:28) Staged Rollouts and Canary Deployments (29:30) The Hundred Megabit Cable Story (34:10) Building Workarounds into Everything (36:11) Hardest Moments in Scaling (39:49) What Makes a Great Robotics Engineer (43:21) ClosingAdditional ResourcesSimbe is hiringBrought to you by Miru, RobotOps Infra for Scaling Teams:

This EpisodeIn Episode 7, we speak with Saman Farid, Co-founder and CEO of Formic. Formic is a Robots as a Service (RaaS) company that owns, deploys, and manages robot fleets inside U.S. manufacturing facilities, with robots actively running in production across more than 100 factories and on track to become the largest independent robot fleet in the country.Before founding Formic, Saman co-founded Comet Labs, an AI-focused VC fund that backed 40+ robotics and automation companies, then joined Baidu Ventures to help run their $600M global AI fund. He's sat on the boards of more than 30 robotics companies, and what he saw from that vantage point is what convinced him to stop investing and start building.We discuss how to align incentives with the RaaS model, how Formic built the three infrastructure pillars — software, operations, and financial — needed to make long-term RaaS, and why pre-deployment scoping is the most underrated part of a successful rollout. Plus, Saman's take on why so many robotics companies get trapped in proof-of-concept purgatory, and what American companies need to learn from China's robotics ecosystem, a question he's uniquely qualified to answer, having grown up in Beijing and spent years investing across both markets.Timestamps(0:06) Introduction (3:25) Why Robots as a Service (7:12) Building the Infrastructure for RaaS (13:26) The Deployment Playbook (18:16) The Importance of Pre-Work and Battle Scars (20:22) 24/7 Support and Monitoring (26:12) Advice for Scaling Your First Fleet (29:17) Why Companies Get Stuck in Proof of Concept Purgatory (33:57) Lessons from China's Robotics Ecosystem (37:52) Closing Advice

In Episode 6, we speak with Ritukar Vijay, Co-founder and CEO of Ottonomy. Ottonomy builds autonomous indoor-outdoor delivery robots for enterprises across healthcare, logistics, aerospace, and more.Before founding Ottonomy, Ritukar spent 16 years in autonomy, leading BMW's L4 self-driving program at Aptiv and building everything from defense UGVs to warehousing robots, with 57+ patents to his name.We discuss why robot ops is really customer ops and most teams get the framing wrong, how Ottonomy templatized infrastructure integrations like elevators and access doors to avoid the custom-engineering trap at every new site, and Ritukar's take that the industry overfits on the robot when the real scaling unlock is the orchestration layer that sits above it. Plus, why he thinks teleoperated robots will hit a wall at scale and full autonomy from day one is the only viable path for enterprise deployments.Timestamps(0:04) Introduction (1:20) What Ottonomy Does (2:53) Healthcare Use Cases (5:37) What Feels Different About Robotics Today (8:01) Educating Customers and the Buyer Process (10:05) Deployment Blockers and Infrastructure Integration (14:36) Shifting from Product Focus to Solution Focus (19:07) Scaling Operations and Robot Ops (23:14) Metrics and Tracking Performance (25:55) Incident Lifecycle and Resolution (27:39) Scaling the Human-to-Robot Ratio (29:50) Indoor-Outdoor Differentiation and Building for the Future (35:09) Advice for Founders and Closing

In Episode 5, we speak with Usman Khan, Founding Engineer at Aigen. Aigen builds solar-powered autonomous rovers for agriculture, starting with mechanical weeding.We discuss why hardware, not software, is the real bottleneck holding back reliable robotics, how Aigen designs to tight constraints from day one instead of falling into the "we'll optimize later" trap that prevents scaling, and Usman's contrarian take that most robotics teams should lower their frame rates. Plus, why he thinks physical AI won't have its ChatGPT moment in 2026, and what will actually bridge the gap instead.Timestamps(0:04) Introduction(0:47) What Aigen Does(3:18) What True Autonomy Really Means(5:59) Hardware Reliability Challenges(10:05) Designing for Reliability(12:43) Testing Practices(14:36) Going from R&D to Production(18:36) Model Deployment and Data Pipelines(23:26) Hot Takes on Data and Design Constraints(28:20) Why Physical AI Won't Have a ChatGPT Moment(32:06) VLAs, Knowledge Distillation, and Edge Compute(35:47) Advice for Scaling Your First FleetSponsorThis episode is brought to you by Miru: Fleet Management for Scaling Robotics Teams.

In Episode 4, we speak with Ashutosh Saxena, CEO and Founder of Torque AGI. Torque builds robotic foundation models for Fortune 100 companies like John Deere.Previously, Ashutosh took a company public and got his PhD under Andrew Ng at Stanford. He's also our first guest with a Wikipedia page!We discuss the gap between demos and production autonomy, why physics-grounded AI reaches reliability faster than pure data approaches, and Ash's contrarian take on building generalist robots: deploy specific models first, let real robots collect edge cases, and iterate fast. Plus, why he thinks the industry is wrong about needing trillions of data points for learned models!Timestamps(00:00) Intro(00:48) About TorqueAGI(01:54) Demo vs. Production Autonomy(03:44) Reliability and Interpretability in Autonomy(06:48) Cross-Domain Learning(10:38) Milestones for Production(12:35) The Importance of Edge Inference(14:27) The Road to Generalist Robots(15:54) How to Collect Data(19:47) The Biggest Bottleneck Today(21:24) Closing

In Episode 3, we speak with Amit Moran, former CTO of Civ Robotics (construction robots for solar) and co-founder of Indoor Robotics (autonomous indoor drones for security). We discuss what it actually takes to scale robot fleets from a CTO's perspective: designing for modularity, setting BOM targets from day one, building self-healing systems, and saving data you don't yet know you'll need.It's the kind of advice you only get from someone who's scaled fleets twice and learned the hard way.𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒(00:04) Intro(01:45) Experience at Indoor and Civ(02:57) Core principles for scaling(6:49) Hardware and Software Modularity(9:45) Infrastructure Rigor vs. Velocity(12:46) Reliability in Autonomy(14:11) Reliability & Fail-Safes(16:00) Self-Healing(18:49) Observability(24:04) BOM & Margins(29:56) Build vs. Buy(31:18) Closing

In episode 2, we speak with Matan Yemini, a seasoned operator with over a decade of experience in drones and industrial automation at teams such as Amazon Robotics/Prime Air and Airobotics.He is now the COO at Rebellion Defense.In this episode, we discuss pragmatic lessons from scaling fleets and delivering real ROI for robotics customers. The insights are often unglamorous, but they require attention to detail and a relentless focus on business value.𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒(0:00) Intro(1:30) Previous Experience(2:21) Defining ROI in robotics(5:12) Shortening payback cycles(7:18) Resources for operators(8:35) Inputs that define ROI(9:55) Evaluating site readiness(12:07) Early signs of a successful deployment(15:29) Getting ground-truth data(18:10) Responding to downtime(20:41) Dealing with regulation(23:05) Responding to downtime(25:28) ClosingThis episode is brought to you by Miru, Config Management for Scaling Robotics Teams.

Joining us in Episode 1 is Joshua Chaitin-Pollak, an industry veteran in robotics. He was formerly:An early software engineer at Kiva Systems, which was later acquired by Amazon, and is now Amazon Robotics. They've now deployed over 1 million robots, making them the largest robot fleet in the world.Software lead at 6 River Systems, acquired by Shopify, and now spun out to Ocado.Software lead at Pickle Robot (Series B)He now leads Software at ATI, a Series A tire-installation robotics company. And they're hiring!In this episode, Josh talks about software operations lessons from scaling a fleet, how to stand up a TechOps team, and what he's focusing on starting from day 0 to make his fleet ready for scale.𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒(00:00) Intro(00:22) The biggest fleet Josh has scaled(01:49) How do you operate large fleets?(03:24) How to forecast when things start breaking(05:59) Best practices for software operations(07:27) Combating process theater(09:27) What is TechOps/RobotOps?(10:21) What backgrounds make strong TechOps hires?(11:51) The relationships between TechOps and other functions(13:25) Biggest bottleneck for software teams in robotics today𝐒𝐏𝐎𝐍𝐒𝐎𝐑𝐒This episode is brought to you by Miru, Config Management for Scaling Robotics Teams.