loading
loading
As India’s physical AI ambitions remain heavily tied to imported robotics hardware, US-based Avathon is taking a different route. It is building the intelligence layer that controls physical systems, while using India’s engineering talent and academic institutions to develop the deeper research needed to make those systems work across industrial environments. The company, which has operated since 2013, has set up a Physical AI Centre of Excellence in Bengaluru, with around 150 employees already working in India, and is partnering with IIT Roorkee on a physical AI research lab. Pervinder Johar, CEO, Avathon tells AIM that the company intends to fund PhDs and pursue research that could take years to translate into commercial products. “Our focus has been on applying AI to make that infrastructure run better,” Johar adds. IIT Roorkee Collaboration The IIT Roorkee partnership is designed around this broader definition of physical AI. The technology cuts across geophysics, civil engineering, aerospace, mechanical engineering, chemistry, physics, and computer science. For example, mining applications can require expertise in seismic and geophysical data, while transportation and aerospace applications bring in civil, mechanical, and aeronautical engineering. “The purpose of the collaboration with IIT Roorkee” is to work on problems that require long-term research, Johar states. Avathon plans to initially sponsor around five PhDs in the first year, although Johar states the company could eventually support 30–40 PhDs a year, depending on the research problems and availability of candidates. The company does not see these as short-term projects. “These things are not the kind of software programming that you want instant gratification,” Johar states. “You have to actually invest for a long, long time to find answers to some of the hardest-to-solve problems.” Avathon is Building the ‘Brain’ Johar draws a distinction between physical AI and the more visible robotics market, where companies often focus on building or modifying robots. Avathon develops AI systems that can control and coordinate physical machines and industrial assets. The company's focus spans energy, mining, oil and gas, transportation, aerospace, and defence, among other industrial sectors. Johar believes the next phase of physical AI will require more than individual intelligent machines. A factory, mine, or industrial site could eventually have autonomous trucks, robots, humanoids, and drones operating simultaneously. “There’s an intelligence within each robot, but then you need intelligence which is going to be almost like an operating system or a control system for all the robots or all the humanoids which are doing a set of tasks,” he says. That broader coordination layer is where Avathon sees its role. The Research Bet One area being explored is asset management. Johar estimates that between $100 trillion and $200 trillion worth of physical infrastructure globally is ageing, creating a problem that cannot simply be solved by replacing old infrastructure. One potential application is using AI to predict equipment or infrastructure failures before they occur and prescribe maintenance. But Avathon sees this as a broader research category rather than a single product. The applications could span renewable energy, traditional energy, nuclear power, electricity distribution, oil and gas refining, and other industrial infrastructure. For Avathon, the value of the IIT partnership is therefore not simply producing another robotics application. It is about developing models and methods for understanding complex physical systems. Avathon Expects Research to Fail The company's approach also highlights one of the biggest differences between conventional AI development and deeptech research: not every experiment needs to produce a product. “We don’t expect every single research project to be successful,” Johar says. “If half the things fail, it’s okay. It’s okay. The other half, as long as they succeed, you’re doing very well. You don’t have to be that 80, 90% of the things are going to be successful.” For Avathon, even unsuccessful research can generate useful knowledge. “As much of research is that something works, it’s also research when you can prove out that something doesn’t work,” Johar says. India’s Physical AI Gap Johar’s confidence in India is not based simply on the size of its AI talent pool. “It is a question of focus and direction.” Avathon already has more than 150 patents, according to Johar, with around 30 additional applications related to physical AI. The company plans to use that existing research base as a starting point for its Indian teams and academic partners. The more difficult question for India is whether an AI talent base alone can support a physical AI industry. Johar argues that it cannot. “Every economy that needs to be successful, you have to have a combination of both manufacturing and computer science,” he says. This is particularly relevant to India, where companies increasingly work with imported robotics hardware and focus on software, AI integration and deployment. Avathon itself conducts some manufacturing work, but primarily in laboratory environments rather than at mass-production scale. “We make it easier to do manufacturing,” Johar says, referring to the company's AI systems. The goal is to reduce errors while maintaining quality. That leaves a larger ecosystem question for India: whether its physical AI ambitions can move beyond integrating imported machines to developing the manufacturing, components, control systems, and research capabilities required to build complete systems domestically. Industrial Ecosystems Johar sees the next challenge in physical AI as coordination, not simply making individual machines smarter. A mine, for example, could eventually have an autonomous truck, a robot, humanoids and drones operating together. The intelligence required would therefore have to span the entire operation. “A lot of the work that we think is still ahead of us is not intelligence of any one of these particular components. It’s the intelligence across the entire ecosystem which needs to work together,” he states. That is also the rationale behind Avathon’s decision not to scale its academic collaborations too quickly. Johar states the company wants to spend the first year validating the IIT Roorkee model before considering expansion to other institutions. For now, Avathon is betting that India can develop the research and engineering base for physical AI—but its own strategy suggests that the country will need more than software talent to close the gap. It will need sustained investment in deeptech research, manufacturing and the physical infrastructure on which AI systems ultimately depend
document fetched 2026-10-24 11:25