With decades of experience navigating the intricate worlds of logistics and supply chain, Rohit Laila has a unique vantage point on the evolution of automation. His passion for technology and innovation offers a grounded yet forward-looking perspective on the industry’s next great leap. In this discussion, we delve into the landmark partnership between Humanoid and Schaeffler, exploring the real-world hurdles of integrating humanoid robots into factory floors. Rohit breaks down the strategic thinking behind their phased deployment model, the critical role of hardware collaboration in accelerating development, and the sophisticated data strategies used to teach these machines complex industrial skills.
Your initial 2026-2027 deployment will validate beta-stage robots against metrics like performance and reliability. What specific operational challenges do you anticipate when integrating these humanoids into Schaeffler’s existing workflows, and how will you quantitatively measure success to prepare for a broader rollout?
The biggest challenge isn’t just about the robot’s capability in isolation; it’s about its fluency within an established, human-centric environment. You’re introducing a new kind of “employee” that doesn’t get tired but also doesn’t have the intuition of a person. I anticipate initial friction in hand-off points between the robot and human workers or other automated systems. The real test is in the grey areas—adapting to slight variations in part placement or navigating a walkway that’s unexpectedly busy. Success will be measured with a very strict scorecard. We’re not just looking at raw performance, but at availability and serviceability. A robot that completes a task quickly but requires constant maintenance or is difficult to integrate into the IT security framework is a net loss. The 2026-2027 phase is about hardening the system against the chaos of a real production floor, ensuring it’s not just a clever machine, but a reliable industrial tool.
The plan is to begin with a Robot-as-a-Service model for the gamma phase before offering a capital expenditure option. What is the strategic advantage of this phased approach, and what key performance indicators will signal that a facility is ready to shift from RaaS to CapEx?
The RaaS-first approach is an incredibly smart way to de-risk a massive technological leap for both parties. For Schaeffler, it lowers the initial financial barrier and operational burden; they’re paying for an outcome, not just a piece of hardware. This allows them to test and validate without a huge upfront investment. For Humanoid, it keeps their team deeply embedded in the operations, ensuring the robots are performing optimally and gathering crucial data during the gamma phase. The shift to a CapEx option will be triggered when the robots become utterly predictable. The key performance indicators will be metrics like availability and reliability hitting consistent, high marks over several quarters. When Schaeffler’s own operations team can forecast the robot’s output and maintenance needs as accurately as any other piece of factory equipment, that’s the signal. It means the technology has matured from an innovative service into a core, dependable asset they can confidently own and manage.
With Schaeffler becoming the preferred supplier for joint actuators, how does this strategic hardware partnership accelerate your development and improve robot performance? Could you detail the technical requirements you are jointly exploring for the next generation of these critical components?
This is the secret sauce of the whole agreement. It’s not just a supply deal; it’s a symbiotic co-development loop. Humanoid doesn’t have to reinvent the wheel on motion technology because they’re partnering with a world leader. This immediately accelerates their R&D, allowing them to focus on the AI and systems integration. Schaeffler’s expertise means the actuators will be more powerful, efficient, and durable than anything Humanoid could develop alone in the same timeframe. I imagine the technical requirements they’re exploring for the next generation are centered on power density, thermal management, and precision. They’ll be asking, “How can we make a joint that’s stronger but lighter? How can we make it run for hours under heavy load without overheating? How can we give it the fine motor control needed for delicate assembly tasks?” This collaboration ensures the robot’s physical body evolves in lockstep with its digital brain.
Data will be gathered via teleoperation and synthetic generation to train AI models for specific industrial skills. How do you balance these two data sources to develop robust capabilities, and can you walk me through the steps from initial data collection to a robot reliably performing a complex task?
It’s a two-pronged approach that’s essential for building a truly intelligent system. Think of synthetic data as the textbook; you can generate millions of simulated scenarios to teach the robot the fundamental physics and geometry of a task, like bin picking. This builds a massive, foundational knowledge base. But the real world is messy. That’s where teleoperation comes in—it’s the hands-on apprenticeship. An operator guides the robot through the task, and the AI learns from the nuance, the unexpected friction of a part, or the specific lighting of the factory. The process starts with a large synthetic dataset to create a baseline AI model. Then, that model is deployed on the robot, and teleoperation is used to collect real-world data, especially for edge cases where the model fails. This data is fed back to refine the model, creating a cycle of continuous improvement until the robot can perform the task autonomously and reliably, even with real-world variability.
After a successful bin-picking proof-of-concept in 2025, you are scaling up to hundreds of robots. What were the key learnings from that initial test that gave both companies the confidence to commit to this multi-year deployment and deep technical collaboration?
That bin-picking POC in 2025 was the “moment of truth.” It was far more than just a technical demonstration. It proved that the core perception and manipulation technology was viable in a practical, industrial context. The key learning wasn’t just “can the robot pick an item from a bin,” but “can it do so with the beginnings of industrial-grade reliability?” I can just imagine the scene—engineers from both companies watching this pre-alpha robot, which is still a raw prototype, successfully identify, grasp, and place a component from a cluttered bin. That single, successful action transformed an ambitious idea into a tangible business case. It demonstrated that the foundational AI and hardware were sound, giving both Humanoid and Schaeffler the confidence that this wasn’t just a science project. It was the critical proof point that justified committing to a five-year plan and the deep integration of their engineering teams.
What is your forecast for the widespread adoption of humanoid robots in industrial manufacturing over the next decade?
I believe the next five years will be defined by targeted, large-scale deployments in controlled environments like we’re seeing with Schaeffler. We’ll move beyond pilots and see hundreds, then thousands, of humanoids performing specific, high-value tasks in logistics and manufacturing. The primary hurdles are no longer just technological; they are about integration, safety certification, and perfecting the economic model, whether it’s RaaS or CapEx. In the latter half of the decade, as the technology matures and costs decrease, we will see adoption accelerate dramatically. The focus will shift from single-task applications to more general-purpose roles, with robots capable of learning new skills on the fly. The conversation is shifting from “if” humanoids will be part of the industrial workforce to “how quickly and deeply” they will be integrated.
