Formant Verified

In this conversation, Anish Shah, CEO of Trossen Robotics, discusses the evolution of Physical AI, the significance of training data, and the impact of the Aloha project on the robotics industry. He highlights Trossen's journey from a traditional robotics manufacturer to a key player in the Physical AI landscape, emphasizing the importance of collaboration and innovation in driving the field forward. This conversation explores the evolving landscape of Physical AI, particularly in biopharma and robotics, highlighting the shift from hype to pragmatism, the changing customer base, and the implications of teleoperation. It delves into the potential for a SaaS-like moment in robotics and the ongoing uncertainty surrounding data quality in the field.

Takeaways

• Physical AI relies on extensive data
• Collaboration drives innovation in robotics Physical AI is gaining traction
• Teleoperation will be a long-term solution

Chapters

• 00:00 Introduction to Trossen Robotics
• 03:00 The Importance of Physical AI Training Data
• 06:00 The Aloha Project and Its Impact
• 08:52 Evolution of Trossen Robotics
• 11:56 Collaboration with Google DeepMind
• 15:01 Current State of Trossen Robotics
• 17:56 Applications of Physical AI
• 20:51 Market Trends and Future Outlook
• 27:19 The Rise of Physical AI in Biopharma
• 28:44 Shifting Mindsets: From Hype to Pragmatism
• 30:42 Evolving Customer Base in Robotics
• 32:56 The Impact of AI Models on Robotics
• 35:44 Market Trends and the Future of Robotics
• 39:33 The SaaS Moment for Robotics
• 46:21 Teleoperation: A Long-Term Solution
• 51:03 Teleoperation and Labor Arbitrage
• 56:27 The Uncertainty of Good Data in Physical AI

In this conversation, Henrik and Gio discuss the evolution of Awign, a company specializing in robotic training data, and the dynamics of the gig workforce involved in data collection. They explore the challenges and opportunities in the robotics data landscape, including customer engagement, data types, and market demand. The conversation delves into the evolving landscape of data collection for robotics, highlighting the maturation of standards, buyer expectations, and the challenges of scaling operations. It also explores the nuances of teleoperation, data quality, and the competitive dynamics between different global markets.

Takeaways

• Awign transforms human work into robot-ready data
• The gig workforce is crucial for data collection Data collection standards are evolving
• Teleoperation presents unique challenges

Chapters

• 00:00 Introduction and Setting the Scene
• 05:09 Understanding Awign's Mission and Offerings
• 11:52 Navigating the Robotics Data Landscape
• 16:37 The Gig Workforce and Its Dynamics
• 21:12 Customer Insights and Market Evolution
• 24:10 Data Types and Market Demand
• 26:29 Evolution of Data Collection Standards
• 28:08 Navigating Buyer Expectations
• 32:06 Scaling Data Collection Programs
• 34:24 Teleoperation and Staffing Challenges
• 36:48 Teleoperation in Robotics
• 38:08 Ensuring Data Quality and Privacy
• 39:44 Understanding Data Diversity
• 41:37 Payment Models in Data Collection
• 43:57 Emerging Standards in Robotics Data
• 45:31 Market Trends in Data Commoditization
• 47:22 India vs. China in Robotics Data
• 49:55 Future Market Predictions
• 52:20 Strategic Positioning in Data Collection
• 55:47 Advice for New Entrants in the Market

The conversation explores the evolving landscape of data infrastructure and AI, focusing on the transition from traditional data engineering to modern AI applications. Ryan Wexler discusses the challenges and opportunities in the data market, particularly in relation to AI training data and the role of data teams in businesses. The conversation delves into the complexities of robotics training data, the current state of the market, and the future of robotics and AI. It highlights the challenges in delivering effective training tasks, the importance of specific data for model training, and the potential for growth in the robotics sector.

Takeaways

• Data is crucial for AI success
• AI tools are shifting business dynamics Robotics training is still evolving
• Specific data is crucial for model success

Chapters

• 00:00 Introduction to Data Infrastructure
• 02:59 The Evolution of Data Engineering
• 06:07 Challenges in the AI Era
• 08:57 Investment Trends in AI Startups
• 11:56 Frontier AI vs. Physical AI
• 15:07 The Economics of Data in Robotics
• 17:53 The Role of Experts in AI Training
• 20:48 Comparing LLMs and Robotics Data
• 24:01 Current State of Robotics Data Market
• 30:14 Navigating the Challenges of Robotics Training
• 34:30 Identifying Buyers in the Robotics Market
• 39:00 The Bitter Lesson in Robotics
• 45:50 The Role of Teleoperation in Robotics
• 51:48 Rethinking Autonomy in Robotics

The conversation explores the evolving landscape of robotics data, focusing on the types of data used, the importance of real-world data collection, and the challenges faced in training robots effectively. Miki, the co-founder of Northstar, discusses the significance of high-quality data, the transition from lab to real-world applications, and the future of robotics in terms of deployment and data collection. The conversation explores the evolving landscape of robotics, focusing on teleoperation, hardware innovations, and the future of data collection in robotics. Miki discusses the importance of diverse data sets, the role of teleoperation in achieving autonomy, and the challenges and opportunities in the robotics data market.

Takeaways

• Importance of real-world data
• Future of robotics deployment Teleoperation is essential for robotics
• Diverse data sets enhance robot performance

Chapters

• 00:00 Introduction to Robotics Data
• 03:08 Current Landscape of Robotics Data
• 05:03 Understanding Trainable Data
• 06:35 Challenges of Lab vs. Real-World Data
• 08:37 Market Dynamics in Robotics Data
• 10:39 Recent Developments in Physical AI
• 12:59 Pricing and Customization of Data
• 15:34 Quality Control in Data Collection
• 17:09 Northstar's Vision for Robotics
• 18:39 Future Data Modalities
• 20:32 Reliability in Robotics
• 21:56 Lessons from Frontier AI
• 23:54 Differences in Physical AI Training Data
• 25:51 Teleoperation and Remote Control
• 29:23 Teleoperation: A Bridge to Autonomy
• 31:50 Innovations in Robotics Hardware
• 35:55 The Future of Data Collection
• 41:15 Data Capture Day: Behind the Scenes
• 47:32 Fair Compensation in Data Collection
• 50:55 The Robotics Data Market Landscape
• 53:12 Understanding Valuable Data in Robotics

The conversation explores the challenges and insights from Sam Padilla's experience in the robotics data market, particularly focusing on the difficulties of building a sustainable business around robotics training data. Sam discusses the complexities of data collection, the importance of understanding buyer needs, and the evolving landscape of data in robotics. The conversation delves into the evolving landscape of robotics and AI, focusing on the challenges and opportunities for newcomers in the industry, the emotional toll of startup failures, and the importance of transparency and community engagement. Sam Padilla shares insights on the future of humanoid robots, the significance of learning from failures, and the potential for embodied AI to revolutionize various sectors.

Takeaways

• Robotics data market is challenging
• Understanding buyer needs is crucial Embrace failure as a learning tool
• Focus on community and transparency

Chapters

• 00:00 Introduction to Robotics Data Challenges
• 02:35 The Complexity of Robotics Data Collection
• 07:13 Go-to-Market Strategies in Robotics Data
• 10:53 The Labor-Intensive Nature of Data Collection
• 11:23 Geographical Choices for Data Collection
• 12:55 Incentives for Data Collectors
• 14:07 The Commoditization of Video Data
• 15:53 Hardware vs. Software in Data Collection
• 18:15 Market Dynamics and Competitive Landscape
• 20:27 Understanding Data Buyers
• 22:16 The Role of Data Brokers
• 24:00 The Elusive Nature of Data Buyers
• 26:56 The Challenges of Exclusivity in Data Deals
• 30:06 The Lifecycle of Data Value
• 32:30 Navigating Buyer Needs and Data Types
• 34:56 The Future of Robotics Data Market
• 42:20 Navigating the Robotics Landscape
• 48:15 The Emotional Toll of Startup Failure
• 52:03 Open Sourcing and Community Engagement
• 58:16 Humanoid Robots: Pros and Cons
• 01:05:10 Key Lessons for Robotics Entrepreneurs
• 01:10:14 Future Directions in Robotics and AI

There may be only 20 people in the world who can say yes to a million-dollar robotics dataset. Maybe 50.

In this episode of Formant Verified, Henrik Bennetsen sits down with Sam Padilla, founder of Eidon AI and author of "Robotics Data Is a Broken Business." They talk through what Sam learned building a robotics data company, why he open-sourced the data, and why he thinks the market for physical AI training data is much smaller and more fragile than most people assume.

In this episode:

• Why the buyer market for robotics data is "tiny, tiny, tiny," and why go-to-market is really about access
• How Eidon's first sales went through data aggregators, not straight to the big labs
• What it takes to produce one usable hour of data, from custom hardware to QC
• The data flywheel, and why Waymo's safety drivers are the model for today's data collection companies
• Why you should avoid exclusivity deals, and how to price one if you do sign
• Why the window for data collection startups may only be one to three years
• Picking a corner in a crowded market, and the difference between a company dying and shutting down

If you're building in physical AI, buying training data, or thinking about starting a data company, this conversation is a candid look at how the market actually works.

Guest: Sam Padilla, founder of Eidon AI
Host: Henrik Bennetsen, CBO at Formant

Formant Verified is the podcast from Formant about the people building the future of physical AI and robotics.

🎧 Subscribe for more conversations with the builders shaping physical AI.

#PhysicalAI #Robotics #AITrainingData

The conversation with Matt Gordon covers his journey from early career to founding A3 Analytics, insights from Toyota Research Institute, challenges in implementing machine learning in manufacturing, lessons from early days at Palantir, and the interplay between forward deployed engineers and software engineers. Matt also discusses the challenges in advanced manufacturing and AI adoption, the role of AI in manufacturing, and the balancing of software and services at Palantir. The conversation highlights the Rice Theory of Culture and its impact on manufacturing, as well as the contextualization of data with LLMs. Finally, Matt shares insights into the role of forward deployed engineers and the concept of Data-driven Kaizen.

Takeaways

• Data-driven Kaizen
• Challenges in Manufacturing
• Holistic vs Analytic Thinking

Chapters

• 00:00 Challenges in Advanced Manufacturing and AI Adoption

The conversation explores the impact of AI technology in the field, focusing on productivity enhancement and the importance of data structuring. It delves into customer stories, time savings, quality of data, and the future vision of AI in the physical world, highlighting the adoption of robotics and drones, customer interaction, and the evolution of field operations.

Takeaways

• AI in the field enhances productivity
• Data structuring is crucial for AI technology

Chapters

• 00:00 Introduction and Mic Check
• 05:27 Time Sheets and Billing
• 12:02 Rich Context and AI Interfaces
• 17:33 Customer Interaction and Scheduling

The conversation covers Formant's services, Caleb's role as a forward deployed engineer, the challenges of operational complexity in manufacturing, and the potential of AI in predictive maintenance. It also explores the collaboration between software and services, as well as the issue of data-rich, action-poor environments in advanced manufacturing.

Takeaways

• The role of a forward deployed engineer involves on-site interaction with customers and the translation of customer needs into product features.
• The potential of AI in predictive maintenance offers a new approach to solving long-standing problems in advanced manufacturing.

Chapters

• 00:00 Introduction to Formant's Services
• 09:24 Operational Complexity in Manufacturing Facilities
• 16:20 Collaboration Between Software and Services
• 22:33 Predictive Maintenance in Manufacturing
• 27:37 Data Rich, Action Poor in Advanced Manufacturing

In this episode, Henrik Bennetsen and Fin Elliott pull back the curtain on the "Data-Rich, Action-Poor" reality of modern manufacturing. While the world hears "automation," the truth on the ground is often a "tsunami of alarms"—sometimes over 9,000 per day—that leave operators drowning in noise.

Henrik and Fin discuss how Agentic AI is changing the game by moving beyond simple dashboards to active investigation. We dive into:

• The 9,000 Alarm Problem: Why a single facility's sensor systems can generate more data than humans can realistically process.

• Investigation Blueprints: How Formant captures "tribal knowledge" to train AI agents as specialized procedural masters for your specific site.

• The Junior Technician at Scale: Real-world results from semiconductor pilots where AI identified over 50% of alarms as false or "shelved," saving massive amounts of human labor.

• Chaos to Clarity: Shifting from reactive "firefighting" to a stable, scalable operational flow