Will CynLr's Bet On Object Intelligence Revolutionise Industrial Robotics?

A
article last month tried to observe the future of human civilisation as Elon Musk sees it. “Humans won’t just live on Mars. They will also never have to work again. Money will be irrelevant. And everything they could ever want will be immediately accessible,” said the article about what Musk calls ‘sustainable abundance’.

The Tesla chief describes it as a post-scarcity society where technologies would be so ubiquitous and so powerful that it would eliminate the need for labour.

Call it fiction or fantasy, truth comes in support. Robot adoption in factories around the world is growing at a breakneck speed. The global average robot density reached a record 162 units per 10,000 employees in 2023 – more than doubling from the seven-year-old figure. And, in sync is growing the $16.89 Bn global industrial robotics market to scale $29.43 Bn by 2029. As various players try to piece together the jigsaw, a whole-new concept of physical AI takes its birth.

It needn’t always be a factory or a laboratory to ideate an industrial robot. It can be derived even from a coconut farm – just the way CynLr did it.

For Gokul NA, the early trigger was an attempt during his college days to build a robot capable of harvesting coconuts. The idea was simple, but the execution was not. Agricultural variability exposed a deeper truth: robots could see, but they could not truly understand or manipulate the physical world.

“It was a journey, it was not just a moment,” Gokul told Inc42.

Unlike industrial robots that operate in controlled indoor environments, outdoor robots – known as collaborative robots or cobots – are exposed to extreme weather conditions. Cobots used in the farm sector operate in harsh weather conditions where they are continuously exposed to abrasive dust, chemically aggressive fertilisers and pesticides, and mechanical stresses from rough terrains.

Early experiments uncovered a structural gap in robotics. The robotic arms were rigid, pre-programmed and lacked the ability to adapt to different environmental conditions. A robotic arm costing upwards of INR 30 Lakh was, in his words, just six motors and three pipes, and unable to handle contextual variability. At the core of the CynLr blueprint for its 2019 launch lay one simple realisation: seeing is not manipulating.

After spending years at National Instruments India, where Gokul specialised in machine vision working across India, Saudi Arabia and Russia, he and CynLr cofounder Nikhil Ramaswamy stepped out in 2015 to test their approach commercially. In three years, they solved around 30 machine vision problems in industrial settings. Most involved guiding machines to pick, orient and place components.

The success validated the sensing layer. But it also made a limitation clear: customising vision stacks for individual use cases was not scalable.

The company began formally productising this thesis in August 2019. The blueprint, Gokul said, hasn’t changed in the last six years. What changed was the scale, the form factor and the capital required to execute it.

The ambition was not to build a better robot arm. It was to build the intelligence layer that allows machines to dynamically pick, align and mate objects in unpredictable industrial environments.

Eye On Vision Helps Map The Journey

CynLr had estimated that it needed $3.5 Mn to kick off in 2019. But, it started with $770,000.

The outbreak of COVID-19 pandemic led to a collapse of global supply chains. Some robotic components took up to 11 months to be delivered and iteration cycles slowed dramatically. Hardware startups operate on long gestation cycles and the pandemic stretched those timelines further.

Capital constraints forced compromises. Instead of building a full-stack system from scratch, CynLr prioritised vision – the least mature layer in robotics – and borrowed commercially available robotic arms.

But that triggered a second problem. Investors read the company wrong, Gokul noted.

Because CynLr used off-the-shelf robotic arms while building proprietary camera systems, it was frequently labelled a system integrator. To counter that perception, the company showcased its camera block as a standalone product. That solved one misunderstanding but created another: it began to be seen as merely a vision company.

“The calibration to understand how the value gets built in this space is very low in the VC space,” Gokul said.

The mismatch ran deeper than branding. Deeptech revenue curves do not resemble SaaS or consumer tech trajectories. Hardware density, customer pilots, certifications and multi-year validation cycles precede meaningful scale. But investors often asked for conventional metrics: ARR, rapid velocity, short sales cycles.

CynLr’s early need to engage Indian customers also exposed structural limitations. Domestic manufacturers were price-sensitive. R&D was often incremental, rather than foundational. As Gokul put it, many companies could not differentiate between foundational innovation and incremental innovation.

Despite the friction, CynLr raised an additional $4.5 Mn, and over time brought its total funding close to $15 Mn. The capital allowed it to move beyond constrained prototypes and engage global customers. As it aims to deploy 300 robotic systems a year by 2028 and break even with gross margins of about 60%, CynLr, aims at raising fresh funding of $75 Mn, according to media reports.

The company truly began operations in 2021 once supply chains stabilised after the dispersal of COVID. Today, it runs a team of around 80 members across India, Switzerland and the US.

The early undercapitalisation cost time. But it also forced architectural discipline.

Cramped Single Arm To Agile Multi-Arm Structure

CynLr’s first systems paired a prototype camera mounted on a single robotic arm.

The limitation was structural. When the arm was seeing, it was not grasping. And, when it was grasping, it was not seeing. Observation and manipulation were sequential, not concurrent.

Gokul used a human analogy to explain the flaw. If a factory worker had only one hand, one eye and could not move their neck or hip, how many tasks could they realistically perform?

The answer shaped CynLr’s evolution.

With more funding, the company moved from a single arm to dual arm, and eventually three-arm configurations. One arm could stabilise or reposition, another could manipulate, and sensing could occur in parallel. “The multi-arm architecture was not spectacle – it was an engineering realism,” he said.

Customers responded positively to this form factor. According to Gokul, the multi-arm system was a lot more realistic, closer to the deployment scenario, compared to humanoid designs. It also allowed factories to reuse existing infrastructure and trained engineers.

CynLr’s core strength today lies in part-to-part mating tasks. Wherever two components must be picked and assembled with variability in orientation, positioning or material behaviour, its object intelligence stack becomes relevant.

Object Intelligence is a robotics capability that enables machines to perceive, understand and manipulate previously unseen objects without prior training. Using advanced vision-based sensing and real-time learning, the robots analyse an object’s geometry, texture, reflectance and grasp possibilities within seconds. This allows machines to adapt their actions on the fly, instead of relying on pre-programmed routines or large training datasets. By learning through interaction and continuous feedback, Object Intelligence enables robots to function reliably in dynamic, unstructured real-world environments.

Applications include high-precision industrial assembly, dynamic pick-and-place, alignment and insertion of plastic and semi-flexible components that return to original shape. In automotive and semiconductor environments, the system is being evaluated for machine tending, guided assembly and multi-action automation within complex equipment lines.

Under the hood, CynLr borrows simulation tools such as MuJoCo and NVIDIA Omniverse , but builds its own AI/ML models, largely reinforcement learning policies, to simulate and stress-test multiple action pathways.

The long-term ambition is modularity. In the next two years, CynLr aims to offer a Lego kind of kit, where arms, sensing modules, controllers and mobile bases can be configured based on industrial needs.

The path from prototype camera to modular physical AI stack has been iterative. But it has stayed faithful to the original thesis: intelligence must be embodied in physics.

The State of Robotics In India

CynLr has its entire customer base based outside India. The company’s annualised revenue stands at $1 Mn, with about $600,000 generated in the past six months. But, revenue is not yet the focus for Gokul.

Semiconductors account for approximately 70% of its current revenue. Automotive interest remains strong and could invert that ratio in the future.

CynLr targets a $16 Bn market by replacing inefficient existing systems with superior object intelligence. The strategy prioritises achieving $40–100 Mn in revenue through phased milestones and early-stage catalysts.

CynLr is engaged with an automobile manufacturer and two of the world’s largest semiconductor equipment manufacturers in phased, two-year pilot programmes. Full-scale deployment requires industrial safety certifications that can take two to three years.

The upside, if the validation succeeds, is substantial.

One of the clients could alone potentially scale to 300-400 deployments annually after 2028. One semiconductor equipment manufacturer operates roughly 6,000 machines globally, each with 100-plus automatable actions.

CynLr is not optimising for vanity growth metrics. It is optimising for industrial credibility.

On the broader trajectory of physical AI, Gokul remained measured. “AGI… is not going to happen anytime soon,” he said. Industrial penetration will be incremental. Complex assembly tasks will automate gradually. But, household robots navigating uncontrolled environments remain distant.

“The problem is understanding the context,” he noted.

CynLr’s bet is that the most meaningful breakthroughs in physical AI will not emerge from humanoid spectacle, but from structured factory floors, where variability is constrained, value density is high, and object intelligence can compound.

In a world chasing general-purpose machines, CynLr is building specialised intelligence. It competes with global players like Veo Robotics that focuses on 3D sensing for collaborative safety, Flexiv that specialises in force-controlled adaptive robots, besides Covariant, Apera AI, and Alphabet’s Intrinsic. All these companies are in the field of high-precision automation in manufacturing.

Deeptech robotic startups like CynLr are building humanoids that can slowly replace humans with intelligence and vision enhancing their abilities to offer a large section of the human populace a good life and a better living.

[Edited by Kumar Chatterjee]

The post Will CynLr’s Bet On Object Intelligence Revolutionise Industrial Robotics? appeared first on Inc42 Media.