Inside India’s quest to build robotic machines that can think and pick up objects

Under the glare of floodlights in Shenzhen, 32 humanoid robots stepped into an octagonal ring, one after another. They threw punches, spun into kicks, fell, got back on their feet and fought on as thousands cheered from the stands. The Ultimate Robot Knockout Legend (URKL), a humanoid fighting league held last week, looked less like a technology demonstration and more like a sporting spectacle. More than 200 teams from at least 10 countries had entered the competition. Universities, including Stanford, UC Berkeley, Tsinghua and Zhejiang, sent teams to test the intelligence that enabled the robots to see, balance, react and make split-second decisions. This was a public demonstration of a new technological race. Around the world, countries are pouring billions into humanoid robots, betting that the next industrial revolution will be led by intelligent machines that can work in factories, warehouses, hospitals and eventually homes. Thousands of kilometres away, inside an unassuming office in Bengaluru, the race looks very different. There is no gleaming robotics laboratory. No glass enclosure separating engineers from machines. Instead, humanoid-sized robotic arms are scattered across ordinary workstations. Engineers move between robots, desks, and laptops. One robot stretches out three arms, while another waits mid-experiment, surrounded by cables, sensors and half-assembled parts. This is the headquarters of CynLr, one of the Indian startups attempting to build the intelligence behind the next generation of robots. Rather than building another industrial robot, the company is attempting something more fundamental: giving machines the ability to understand and manipulate the physical world the way humans do. Inspired by how infants learn through interaction, CynLr is building what it calls a cognitive layer for robots, a software that allows machines to perceive and handle objects they have never encountered before, without relying on millions of pre-labelled images. But, while global competitors race ahead with billion-dollar investments, India faces a more fundamental question: can it build the artificial intelligence brains of the future without the industrial and manufacturing backbone that has powered robotics leaders such as China, Germany, Japan and the US? Pushing the envelope Humans instinctively judge the shape, weight, texture and orientation of an object before deciding how to pick it up. A robot, however, has to calculate every one of those variables. That is why seemingly simple tasks, placing a bottle cap on a bottle, or picking up a crumpled piece of cloth, remain some of the hardest problems in robotics. That is also the problem CynLr is trying to solve. Rather than training robots on millions of images, its software analyses an object’s geometry, texture, reflections and possible grasp points, allowing the robot to decide how to manipulate it without prior training. “The goal is to make robots learn through interaction, much like a child explores an unfamiliar object, instead of relying on massive labelled datasets,” says Gokul N.A., co-founder overseeing design, product and brand at CynLr. That ambition has also forced the company to work across the hardware-software divide. Today, CynLr develops most of its AI on commercially available robot arms but those arms were not designed for the kind of adaptive behaviour the company wants to achieve. As a result, CynLr has increasingly begun customising its own control systems and components, though its main product is robotic intelligence. Meanwhile, Noida-headquartered and Reliance Industries-backed Addverb is taking a different route. Earlier this year, it unveiled Elixis-W, a wheeled humanoid robot designed for factories and warehouses. Addverb is trying to solve the broader engineering challenge of humanoids, combining mobility, manipulation, perception and autonomy into a single machine that can move around on its own, understand its surroundings, pick up and handle objects, and safely work alongside people. The company develops about 95% of its software and roughly two-thirds of its hardware in-house, with plans to own almost the entire hardware stack over the next three years, according to Prateek Jain, co-founder and chief operations officer. It is also in the process of building a fully walking humanoid. Other startups are approaching the problem from a different angle. Bengaluru-based Mowito is building AI software that teaches existing industrial robotic arms new assembly tasks from as little as a single human demonstration. Instead of spending days programming robots for every production change, manufacturers could teach a new task in minutes. The software combines vision with force sensing, allowing robots to adjust to small variations in parts, lighting and positioning that would normally require reprogramming. The company’s first deployment ran on a line producing up to 2,800 parts a day, where the robot had to complete the task at close to 100% reliability to keep pace. “We built our Vision-Force-Action model from scratch to hit that bar. Across 16 months of continuous operation, the robot has never stopped because of a model failure. Every issue we’ve logged traced back to hardware, operator procedure, or the environment—never the AI,” Puru Rastogi, Mowito’s cofounder and CEO, told Mint. Chennai-based BotForge Labs is pursuing a similar vision of adaptive automation, integrating vision, manipulation and autonomous decision-making so industrial robots can continue performing repetitive tasks even when the environment changes. Together, these companies represent three distinct bets on the future of physical AI. CynLr is building the robot’s brain. Addverb is engineering the complete humanoid. Mowito and BotForge are making existing industrial robots far more adaptive. The different approaches all aim at moving beyond rigid automation towards machines that can understand and interact with the physical world. The arm problem Building a robotic arm is one of the most difficult problems in robotics. A human arm, including the hand and fingers, has more than 20 degrees of freedom (the number of different ways a joint can move). It is powered by muscles and tendons that constantly adjust force, stiffness and position. We instinctively loosen up while holding a fragile glass, tighten our grip if it begins to slip, or change the angle of our wrist to reach around an obstacle. Replicating those seemingly effortless actions in robots requires a complex combination of lightweight material, high-torque actuators, precision gearboxes, force and torque sensors, compliant joints and sophisticated control software. Every joint must know not only where it is, but how much force it is applying and how the environment is pushing back. The robot’s arm is therefore not just an actuator, it is part of its sensing system, continuously gathering information from touch, resistance and motion as it interacts with the world. This is also why the arm has become a strategic bottleneck in robotics. Humanoid robots need actuators that are lighter, more power-efficient and capable of fluid, tendon-like movements while remaining precise enough to assemble electronics or handle delicate objects. A single joint can take years of engineering to perfect, and manufacturing hundreds of identical joints that behave consistently is an even greater challenge. The missing middle Countries that dominate robotics today did not build successful companies overnight—they first set up research ecosystems where universities, manufacturers and startups worked together over decades, say industry experts. In Germany and Japan, large industrial companies routinely fund university research, co-develop technologies with startups, and become early testing grounds for products long before they reach the market. “An engineering PhD in Europe rarely happens without an industry sponsor. That's how research stays connected to real problems. Otherwise you're just writing papers,” says Gokul. In India, academic research and industrial innovation often operate in parallel instead of together, say observers. Universities optimize for publications while companies optimize for products. The gap is visible across manufacturing. Outside pharmaceuticals, where companies routinely partner with research laboratories on drug discovery and development, similar collaborations remain rare in automotive manufacturing, industrial automation and electronics. “How many Indian companies building core manufacturing infrastructure like Siemens do we really have,” asks Gokul. Researchers agree that the gap exists, although they say it is beginning to narrow. “Indian companies invest more in collaborative research today than they did a decade ago, but it is still far behind countries such as Germany, Japan and South Korea,” Prof. Dayal Ramakrushna Parhi and Prof. Ratnakar Dash of the National Institute of Technology, Rourkela said in a joint written response to Mint’s queries. “The bigger issue is synchronisation. Industry, academia and government still don't work together closely enough.” Companies often see university partnerships as a cost rather than a strategic investment, the dons wrote. Academic research typically works on longer timelines, while businesses want commercial outcomes within months. - $42.1 million | Funding raised by Indian robotics startups in the first half of 2026, nearly double the amount raised in H1 2025. - < 1% | Funding raised by Indian robotics startups as a share of that secured by their US peers in 2025. - 56% | Share of publicly listed humanoid robotics companies globally that are Chinese. Prof. Prabhu Rajagopal, director-in-charge and dean (School of Engineering and Science) at the Indian Institute of Technology, Madras’ campus in Zanzibar, Tanzania, points to incentives as one of the missing pieces. “Tax credits, research-linked incentives and preferential government contracts for R&D-intensive companies can encourage industry to invest much more in collaborative research,” he says. Rajagopal’s own lab has established the National Consortium for Non-Destructive Evaluation, bringing together industry partners to jointly develop new technologies. For startups building frontier robotics, the absence of industrial partnerships slows everything down. “We would like manufacturers to get involved much earlier in product development rather than waiting until technologies are fully commercialized. Working together on real industrial problems helps refine products faster, accelerate adoption and build local supply chains for critical components,” says Addverb’s Jain. Pooja Ravishankar, founder of robotics startup Botloka, noted it becomes particularly important in areas such as mould making, printed circuit board or PCB fabrication, electronics assembly and precision manufacturing, critical capabilities for robotics that remain underdeveloped in India, especially for small-batch prototyping. Funding the future Money is finally flowing into robotics. Over the past two years, globally, investors have poured millions into startups building everything from warehouse robots to humanoids and industrial AI. India has also seen a similar sentiment and funding has almost doubled to $42.1 million in the first six months of 2026 from $22.7 million in the comparable span last year, according to research platform Tracxn. But the numbers seem very small when compared to countries globally. Global robotics startups raised over $6 billion in 2025, exceeding the pace of the previous year, according to Crunchbase. Much of the capital went into US companies. In the US, Uber founder Travis Kalanick’s robotics startup Atoms raised $1.7 billion this month from venture capital firm a16z and others. Humanoid startup Figure AI has raised more than $1.5 billion from various investors, including Microsoft and Nvidia. Another humanoid company, Apptronik, has raised over $900 million. Chinese humanoid robotics startups raised 2 billion yuan (about $276 million) across 20 funding deals in just the first two months of 2025. Morgan Stanley estimates that 56% of the publicly listed companies globally involved in humanoid robotics are Chinese. In 2025, India’s robotics startups raised less than 1% of the capital secured by their US peers and about 2% of the amount raised in China, according to Tracxn. Funding has started picking up, but Indian robotics startups are still raising a fraction of what their global peers secure. Robotics startups in the US routinely raise seed rounds of $50–60 million, while the average cheque size in India remains $1–3 million. Even after adjusting for lower costs in India, founders say domestic startups struggle to raise enough capital to build multiple prototypes, experiment with suppliers, or move quickly through engineering cycles. Moreover, founders argue that capital alone won’t determine whether India produces globally competitive robotics companies. The bigger question is whether the ecosystem understands what it is funding. Unlike software startups, robotics companies often spend years solving scientific and engineering problems before meaningful revenue arrives. A breakthrough may come not from acquiring customers, but from getting a robot to grasp an unfamiliar object, improve reliability from 90% to 99.9%, or cut the cost of a critical component in half. Yet many founders say they are still evaluated using SaaS playbooks. “The financial investor cannot be the only voice in the boardroom. You also need people who understand technology and can judge progress using the right metrics,” says Cynlr’s Gokul. Founders say the kind of capital matters as much as the amount. A case in point is Addverb, which says traditional venture investors were hesitant in its early years because of the long development cycles. Strategic investors with a longer time horizon allowed the company to continue investing in research and development while expanding its product portfolio. Vipul Patel, partner at IIMA Ventures, the startup incubator established by the Indian Institute of Management, Ahmedabad, agrees that seed funding for robotics startups is increasingly available, while growth capital follows once a company has revenue and proven economics. The real gap lies in the middle—18 to 36 months after a prototype is built but before it becomes a commercially deployable product. “India has fewer domestic funds with deep hardware expertise and the patience required for this stage,” Patel says. As a result, many startups depend on foreign investors or strategic corporate partners to bridge the gap. Some investors believe the biggest near-term opportunity may not be in robots but in the intelligence that powers them. AI software layers for perception, autonomy and robot control require far less capital than building entire machines, while industrial robotics remains a safer commercial bet than humanoids, which are still likely years away from large-scale deployment. For founders, building robots isn’t simply about writing software or assembling hardware. It requires investors willing to finance years of experimentation before the first commercial success arrives. As Gokul puts it, “The VC who figures out how to evaluate deep-tech properly will become the next transformational investor.” Brains need bodies Even if India succeeds in building world-class robotic intelligence, founders and researchers say that is only half the battle. A robot is ultimately a collection of thousands of precision-engineered parts: motors, gearboxes, sensors, actuators, cameras, bearings, controllers and chips. If most of those components have to be imported, building robots at scale becomes expensive, slower and strategically vulnerable. India can build world-class AI-powered robots, but not yet at scale, wrote Prof. Parhi and Prof. Dash of NIT Rourkela. While the country has clear strengths, a deep AI talent pool, a thriving software ecosystem and a growing startup base, it still relies heavily on imported high-precision sensors, motors and manufacturing equipment, they pointed out, noting that a successful robotics industry needs a strong domestic supply chain with far less dependence on other countries. For companies building robots, the missing components or supplier infrastructure are frustrating. While demand, talent and customer confidence have improved significantly, the next phase of India’s robotics journey will depend on localising more of the hardware stack. “The ecosystem has come of age. Now, localisation of key technologies will determine the next stage of growth,” says Jain. Without an industrial backbone, India’s robotics startups may build the brains, but remain dependent on the rest of the world for the body. Read Mint Long Story for more such in-depth reportage.