From Digital AI to Physical AI: Turning Investment into Real-World Automation on the Warehouse Floor

Rahul Nambiar, Co-founder & Chief Executive Officer (CEO), Botsync

Rahul Nambiar is the Co-founder and Chief Executive Officer of Botsync, a robotics company transforming factories and warehouses through autonomous mobile robots (AMRs) and no-code and vendor-agnostic orchestration software. He leads the company’s strategy, product direction, marketing, and finance, driving Botsync’s mission to make automation more accessible, flexible, and scalable for global enterprises.

 Rahul’s journey into robotics entrepreneurship began at Nanyang Technological University (NTU), Singapore, where he met his co-founders during a robotics competition — an experience that shaped his belief that practical, deployable robotics can solve real industrial challenges. Under his leadership, Botsync has developed the MAG Series of autonomous mobile robots and SyncOS™, a proprietary no-code platform that delivers real-time operational visibility and control, enabling enterprises to streamline material movement and optimize supply chain performance.

Today, Botsync has deployed 200+ robotic systems across 30+ global clients in India, Australia, Thailand, Singapore, Indonesia, and South Africa, with expansion into the United States underway. The company works with leading brands such as Ford, Kimberly-Clark, and Coca-Cola, supporting their transition toward safer, more efficient, and data-driven intralogistics operations.

Recognized as a rising leader in robotics and automation, Rahul was featured in Forbes 30 Under 30 and is a frequent speaker at industry conferences and business forums, where he shares insights on robotics adoption, operational transformation, and the future of industrial automation.

Rahul holds a Bachelor’s degree in Mechatronics from Nanyang Technological University, with a Minor in Entrepreneurship.

1. To start off, could you share a bit about your journey and what led you to co-found Botsync?

My co-founders and I met as students at NTU, where we worked on various robotics projects including the development of autonomous systems. Our early products were catered to roboticists and students to support prototyping and research. Over time, we identified a significant gap in the market for Autonomous Mobile Robots built specifically for industrial material movement, and began developing our MAG line of AMRs.

But as hardware vendors, we quickly ran into a problem that had nothing to do with the robots themselves. Factories and warehouses naturally source machines from multiple vendors based on cost and specialisation, and we found that these systems, despite working toward the same operational goal, simply didn't have a common language with which to communicate and collaborate. Integration was broken, and orchestration was fragmented. Every time a client's workflow needed an update, our specialists had to dig into multiple codebases and effectively start over.

To fix this, we built SyncOS™, a no-code, vendor-agnostic automation control platform that manages mixed robotic fleets, integrates every automation system in a facility, collects data from every stage of operations, and uses AI to make that data actionable. The result is a factory that functions as one cohesive unit rather than a collection of isolated systems. What started as a tool we built for ourselves has become what we believe is the most important layer in modern industrial automation.

2. We’ve been hearing a lot about AI moving beyond software into the physical world. From your perspective, what does this shift really mean for manufacturers on the ground?

For years, AI was essentially a software story — organising recommendations, predicting demand, streamlining back-office processes. That's valuable, but it doesn't move a pallet. What we're seeing now is AI crossing the threshold from screen to physical environment, and that changes everything for manufacturers.

The core distinction is this: digital AI is fed decades of internet-generated text, images, and interactions. Physical AI has to learn from the real world; from sensors, cameras, and LiDAR that perceive objects, edges, textures, spatial relationships, and constantly shifting environmental conditions. That data is far harder to collect, and far harder to make useful.

On the ground, this means machines that don't just process instructions but perceive, adapt, and act in real time. Our AMRs navigate dynamic environments alongside human workers, replan routes autonomously when conditions change, and operate across complex facility layouts without requiring extensive infrastructure changes. But the real shift for manufacturers isn't just about smarter individual robots, it's about what becomes possible when you can collect, centralise, and act on data from every machine in a facility simultaneously. That's where Physical AI moves from an interesting capability to a genuine competitive lever. And that's precisely what we've been building toward with SyncOS™.

3. You’ve pointed to the warehouse floor as the next frontier for AI. What’s driving that shift now, especially with the level of AI investment we’re seeing across markets like Singapore?

Several forces are converging at once. Labour shortages across logistics and manufacturing aren't going away. If anything, they're intensifying structurally across Southeast Asia. Simultaneously, pressure to move goods faster and more reliably has never been greater, and manual operations simply cannot scale to meet demand.

Singapore is a particularly interesting lens for this. The government has been deliberate about channeling AI investment into productivity through programmes like the National Robotics Programme and broader Smart Industry initiatives. That creates an ecosystem effect: access to capital, regulatory openness to deploy robots in real environments, and enterprises that are genuinely willing to pilot. At Botsync, we see ourselves as active contributors to that national agenda by helping translate government ambition into operational reality on the ground, one facility at a time.

The warehouse and factory floor sit right at the intersection of where that investment converts into measurable throughput gains most quickly. But I'd add one thing the investment conversation often misses: data infrastructure. The warehouse floor generates enormous amounts of operational data from every machine, every stage of production, every shift — and almost none of it is being used effectively today. Most factories are fragmented, with machines from multiple vendors running in data silos with no direct way to centralise or act on what they're producing. Solving that problem is what unlocks the real value of AI investment in this space. That's the opportunity we're building for.

4. There’s strong investment going into AI today, but translating that into real productivity is another challenge. Where are we actually seeing measurable impact on the warehouse floor?

The honest answer is: where the use case is specific and the implementation is disciplined. Industry data shows that between 68% and 95% of industrial AI and robotics pilots stall before reaching production, and as few as 4% yield meaningful scaled business value. That's a sobering number, and it tells you something important — the technology isn't the bottleneck. The data infrastructure and integration layer are.

The places where we see genuine, sustained productivity impact are in repetitive, high-volume workflows, which is exactly where our AMRs are deployed. We've seen facilities cut internal material transport time by 50- 60% after deploying our robots, and that's sustained operational performance, not a controlled pilot figure.

What underpins all of this is SyncOS™, our fleet management and orchestration platform which gives operations teams real-time visibility and control across the entire facility, integrating with existing warehouse systems rather than requiring a wholesale replacement. The companies that convert AI investment into real productivity treat it as an operations transformation with a clearly defined problem to solve. The technology performs best when the business problem leads, not the other way around.

5. Can you share a real deployment example — what kind of productivity gains or cost savings are companies actually seeing?

Across our deployments, we’re seeing very tangible, measurable outcomes for customers. In one automotive manufacturing setup, for example, we automated internal material movement using our AMRs, which led to about a 12.5% increase in productivity and a noticeable improvement in throughput, while also reducing reliance on forklifts and manual handling. In logistics, a deployment in Singapore helped save roughly 90 man-hours per month, allowing the team to redeploy labour to higher-value tasks without increasing headcount. More broadly, what we consistently see is improved uptime, more predictable operations, and faster workflows.

What makes these results durable rather than just impressive in the first month is the data layer. SyncOS™ continuously collects operational data from every machine at every stage, and that data compounds over time, enabling smarter decisions about charging schedules, routing, task allocation, and workflow sequencing. For most of our customers, this translates into ROI within two years, and in some cases up to a 30% reduction in operating costs driven by better utilisation of both people and infrastructure.

6. Compared to digital AI, working in a real-world environment is far more complex. What are some of the biggest challenges companies tend to underestimate?

A warehouse isn't a controlled test environment — it's a living, breathing operation. Layouts change. Seasonal inventory shifts the entire floor plan. Workers take shortcuts that no one documents. Lighting conditions vary. New equipment gets introduced. This is something we've engineered Botsync's AMRs specifically to handle; our navigation stack is built for dynamic environments, not just static mapped spaces, so the robots adapt as the facility evolves rather than breaking down every time something shifts.

The second underestimated challenge is the the integration layer. Deploying robots is actually the simpler part. Connecting them meaningfully to your WMS, your ERP, or your conveyor systems, that's where projects stall. SyncOS™ is designed with this in mind, offering integration interfaces that work with legacy systems rather than demanding infrastructure overhauls.

Finally, there's change management, which almost everyone underestimates. Your workforce has to trust these systems enough to work alongside them productively. That takes deliberate effort and something we actively support customers through during deployment, because even the best robot fleet underperforms if the people on the floor aren't genuinely bought in.

7. A big part of this comes down to data from physical environments. How difficult is it to train AI to deal with navigation, object detection, and constantly changing conditions?

Genuinely hard, and consistently underestimated by anyone who approaches it from a digital AI background. The fundamental difference is this: a language model gets trained on decades of internet-generated content that is abundant, structured, and largely pre-labelled by human intent. A robot navigating a warehouse floor learns from camera frames, depth signals, LiDAR point clouds, and spatial annotations that reflect the exact environment it will operate in. You cannot scrape a factory floor the way you scrape the internet.

This creates two problems that compound on each other. The first is coverage. Real-world data collection is never exhaustive. You might run normal operations for weeks and still never capture the edge cases that reality will eventually serve up — glare on a polished floor, a pallet sticking out by 12 centimetres, a worker stepping into a path at the wrong moment, a conveyor stopping unexpectedly mid-shift. These are precisely the situations where systems fail, and they're the hardest to collect data for at scale.

The second is environmental variability. A navigation model trained on a static facility map will degrade the moment that facility changes, and facilities change constantly. Layouts shift seasonally. New equipment gets added. Workers create informal pathways that no one documents. Our AMRs are built with SLAM-based navigation specifically so they don't rely on fixed maps — they build and continuously update their understanding of the environment as it evolves, rather than breaking when it deviates from what they were trained on.

The honest answer to how we address the data problem more broadly is interoperability. By connecting every machine in a facility through SyncOS™ — AMRs, conveyor systems, robotic arms, sensors across every vendor — we collect production data across every stage of operations simultaneously. That gives us a training dataset that is not just larger but genuinely more diverse: data from machines that are wheeled, tracked, and stationary; from environments that are dynamic and varied; from edge cases that accumulate naturally over real deployments. An AI model trained on data from that kind of interconnected, multi-vendor environment will always be more robust than one trained on a single robot type in a single facility. That's the data flywheel that SyncOS™ is designed to build — and it compounds in value with every deployment we run.

8. Many facilities are still running on legacy systems. How realistic is it for them to adopt Physical AI without major disruption?

More realistic than most people assume, but only if the approach is right. The common mistake is treating automation as an infrastructure replacement project. Most brownfield facilities contain a patchwork of older controls, legacy machines, partial upgrades, and disconnected software platforms that were never designed to communicate with each other. Trying to rip and replace all of that at once is exactly how projects stall.

The smarter path is a software-first, integration-led approach. Rather than requiring legacy systems to be replaced, you build a layer above them that enables them to communicate and share data — which is fundamentally what SyncOS™ does. Our platform connects to existing PLCs, conveyors, and legacy systems through standard integration interfaces, making them legible to the broader automation ecosystem without requiring you to touch the underlying hardware. We've deployed in facilities where some machines are decades old, and the interoperability layer handles the translation.

What this means practically is that manufacturers don't have to choose between modernising and maintaining operations. You can start with a single workflow, a specific bottleneck, a handful of robots — prove the value, collect the data, and expand from there. The key insight from our deployments is that the data you collect from even a modest initial deployment compounds in value over time. The facility learns, the models improve, and the next workflow you automate benefits from everything you've already observed. Legacy doesn't have to mean stuck. It just means the integration layer matters even more.

9. If a company decides to take a “wait and see” approach over the next few years, what risks are they really taking?

The competitive gap between early movers and laggards in warehouse automation is already widening, and I see this directly through the customers we work with. The companies that deployed AMRs two or three years ago aren't just ahead on technology — they've built institutional knowledge, trained their people, refined their workflows, and accumulated operational data that continues to make their systems smarter. That learning compounds in ways that are very difficult to close quickly.

There's also a structural cost dimension that's often underestimated. Labour availability is declining in many markets, and the longer you defer, the more exposed your operation becomes to that reality. A "wait and see" position on automation is becoming increasingly difficult to defend to customers who need predictable, scalable operations from their partners.

And perhaps most importantly: the data advantage. Every month an early mover's system runs, it generates more operational data, surfaces more patterns, and enables smarter decisions. That data flywheel is one of the hardest things to catch up on once you're behind. We consistently tell prospective customers that starting doesn't have to mean going all-in. Our modular deployment approach lets you begin with a single workflow, a defined ROI target, and a handful of robots, and scale from there. Waiting isn't neutral. It's a strategic choice with real and compounding consequences.

10. Looking ahead, how do you see warehouses evolving over the next five to ten years as Physical AI becomes more embedded in operations?

The defining shift will be from automated warehouses to truly adaptive ones. Today, even the most sophisticated automated facilities are largely optimised for a fixed operating model. The warehouse of 2030 will reconfigure itself dynamically — rerouting workflows, redeploying robot fleets, adjusting strategies in real time based on order mix, staffing levels, and live supply chain signals.

We'll also see the data advantage become decisive. As interoperability becomes standard and production data flows freely across every machine in a facility, AI won't just surface patterns after the fact, it will predict anomalies before they occur, recommend process optimisations dynamically, and enable manufacturers to move from reactive decision-making to proactive optimisation. The prompts will get simpler even as the intelligence behind them grows more sophisticated. Instead of scraping through multiple log files to understand a bottleneck, an operations manager will simply ask "what's limiting my production capacity right now?" and get an answer with recommended actions in seconds.

The human role will evolve significantly too. It won't disappear, but it will shift toward exception handling, oversight, and the contextual judgement that Physical AI still struggles with. The workers who thrive will be those who learn to collaborate with these systems — and part of our responsibility as a company is making that collaboration intuitive rather than intimidating.

At a macro level, I believe Physical AI will start to redraw some of the economics that historically drove offshoring. When you can automate a significant portion of your warehouse and manufacturing operations, proximity to the customer and supply chain resilience start to outweigh pure labour arbitrage. For markets like Singapore and for companies like Botsync building this technology here, that's not just an industry trend. It's a genuine and exciting strategic opportunity.