Physical AI is the next frontier, and we are taking up the challenge: Raghu Dharmaraju, CEO, ARTPARK | Technology News

Venkatesh Kannaiah: Inform us about ARTPARK, its historical past and its journey.

Raghu Dharmaraju: The Nationwide Mission on Interdisciplinary Cyber-Bodily Techniques created round 25 hubs to concentrate on numerous thematic areas, and we had been recognized as one of many prime hubs for robotics and AI methods.

We started our actions in 2020 and thus far have invested in about 30 corporations. The very first set of startups, about 5 – 6, has already attracted follow-on investments.

We aren’t a typical incubator or accelerator. We contemplate ourselves as an innovation and enterprise builder. It means we frequently begin very early, generally earlier than a full group is shaped, and even earlier than there’s a group in any respect.

We assist deliver such groups collectively to go after particular issues or ideas which will have been initiated by a technologist, typically as an innovator-in-residence. Over time, this turns into an innovation undertaking, which then evolves right into a for-profit or non-profit enterprise. Whether it is for-profit, it goes out into the world as a startup. Whether it is non-profit, it could proceed beneath our umbrella as a digital public good. In all circumstances, there’s a group and a transparent articulation of that group.

We recognise that this sort of deep-tech journey requires sustained help, and we help it with grants.

Venkatesh Kannaiah: Do these researchers, startups or college students come to you from IISC Bengaluru or from all throughout India?

Raghu Dharmaraju: Solely about one-third of our funnel comes from IISc. And the remainder comes from throughout India. After we discuss a funnel, there are actually a number of funnels. There may be an concepts or ideas funnel, and there may be additionally a expertise funnel.

We sit on the assembly level of those. Typically we begin with the expertise, they usually have an idea that step by step takes form and ultimately turns into a startup. At different instances, there’s a identified want or downside, and we’re lucky sufficient to establish innovators-in-residence who can take that ahead.

So it isn’t at all times very linear, that there’s a person with a totally shaped idea and know-how, and every thing proceeds easily from there. That not often occurs.

Venkatesh Kannaiah: The place do you get these downside statements, and do you get them from authorities departments too?

Raghu Dharmaraju: Typically the concepts come from the innovators themselves, and at different instances, they arrive from trade.

In sure circumstances, particularly in defence, they could come from the federal government. Equally, in areas like AI for social impression, the challenges typically come up from public sector wants, akin to public well being points or recurring illness outbreaks.

Venkatesh Kannaiah: Inform us about your programmes and what they search to attain.

Raghu Dharmaraju: Our programmes are what we name innovation programmes. Usually, tasks begin at round Know-how Readiness Degree 3 or 4, and we goal to take them to Know-how Readiness Degree 6 or 7.

This includes a mix of help mechanisms. One is offering amenities. We’ve got amenities for robotics and AI to prototype, take a look at, and even do small-scale manufacturing. We’re additionally a 5G testbed. For instance, now we have a small GPU cluster that groups can use. As well as, being a stone’s throw away from the Peenya Industrial space permits us to do sure sorts of prototyping in a short time.

The second, and maybe crucial, is enabling collaborations with potential clients. We name this the co-creation effort. This occurs via mentorship networks and structured collaborations.

If you wish to do troublesome issues, it isn’t going to occur with Rs 50 lakh and even Rs one crore. These efforts require a number of crores of what you would possibly name pre-commercial R&D. Earlier than business gamers are keen to place cash in, now we have to help the work considerably.

Venkatesh Kannaiah: Are you able to inform us about a few of your startups and the issues they’re fixing?

Raghu Dharmaraju: Ours is tech for good — each for-profit and non-profit. I’ll begin with the for-profit facet.

One among our startups is Twara Robotics, which makes robotic arms for manufacturing. They’ve designed, developed, and constructed robotics parts akin to actuators and tender grippers. Additionally they went on to design, develop, and construct full robotic arms, attracting funding from reputed trade gamers.

There may be Comrado Aerospace. They’ve constructed a hybrid vertical take-off and touchdown drone with a variety of over 100 kilometres and an endurance of about six hours. It’s helpful for surveillance purposes. This is a pretty big drone. It’s now in pilot deployment with the defence forces.

There may be ZenteIQ, which is constructing what are referred to as scientific basis fashions, that are foundational AI fashions for engineering design. Consider purposes like thermal evaluation or computational fluid dynamics. These sorts of instruments are used throughout engineering domains — whether or not you might be designing a constructing, an plane, or different advanced methods. ZenteIQ is likely one of the groups that received the IndiaAI Mission’s basis mannequin problem.

There may be Qosmic Labs, which is creating tech for floor stations and satellite tv for pc communications. Usually, satellite tv for pc communications depend on radio frequencies, that are a comparatively slim pipe. This firm has developed an optical, laser-based communications know-how that will increase throughput many-fold.

There may be FLO Mobility, which has constructed automated construction-site AMRs or autonomous cellular robots for materials dealing with. Consider it as a small truck or trolley. You load the fabric, and it robotically navigates giant development websites. It runs on wheels and might use the short-term lifts which can be put in at development websites to maneuver between flooring.

There may be DexSent Robotics, began by a latest PhD graduate from IIT Gandhinagar. They’re engaged on a dexterous gripper. Consider it as a three-fingered hand that may be connected to a robotic arm. With three fingers, you are able to do fairly attention-grabbing issues — you’ll be able to choose up a cherry, a flask, or many different objects. There are numerous duties on manufacturing flooring that require this stage of dexterity, however the place people can’t be current 24 hours a day.

Venkatesh Kannaiah: Inform us about your improvements/startups that are fixing attention-grabbing social issues.

Raghu Dharmaraju: On the social impression facet, now we have labored on constructing digital assistants for frontline well being employees. Consider this as a WhatsApp-based chatbot.

It’s multi-modal and multi-lingual — multi-modal within the sense that it helps each speech and textual content. It could deal with Hindi in addition to native languages and dialectical variations. This can be a non-profit effort beneath our umbrella. Authorities programmes are utilizing it in locations like jap Uttar Pradesh.

We’ve got additionally been funded by Google, via IISc, to create giant open-source datasets, particularly for speech. One such effort known as Venture Vaani,an initiative we constructed from scratch in collaboration with Professor Prashanth Ghosh at IISc.

By means of this effort, now we have found points of our cultural heritage, together with meals objects and names we could not have heard of in any other case.

Vaani is a non-task-specific dataset. It’s utilized by numerous teams to enhance their fashions and might match into any conversational AI system, together with authorities programmes for conversational AI and automated speech recognition. Analysis has proven that there are actual district-to-district variations in mannequin efficiency. With out that variety within the information, the fashions merely don’t study.

One other instance of our work is a dengue outbreak prediction system being developed and piloted in partnership with the Authorities of Karnataka. It’s operational as a pilot. An vital side of this work is that these fashions are by no means static. You can’t freeze them. Local weather patterns, mobility patterns, and different components maintain altering, so the fashions must be up to date each season.

We’re additionally engaged on heat-health danger fashions. In the summertime of 2024, for instance, we noticed heatwave-related deaths being formally recorded in giant numbers for the primary time. Temperature alone just isn’t sufficient; you additionally must consider humidity, wind, and different variables to know the precise well being danger. We’re working in the direction of rising the granularity of those predictions and are additionally aiming for a seven- to ten-day forecast window.

We additionally work on information requirements and benchmarking necessities. For example, if somebody claims to have an AI system for deciphering X-rays for TB, how do you benchmark it? How are you aware it’s adequate to deploy? We work on defining the information requirements and necessities wanted for that.

We’ve got additionally created a platform for environmental surveillance. It permits a pattern to be tracked finish to finish — from the purpose the place it’s collected from the atmosphere, whether or not that’s soil, milk, water, or the rest, via testing for pathogens or different indicators, and eventually via sequencing.

These efforts span a number of verticals and are pushed by stakeholder wants.

As for the way it’s used, these platforms are usually not meant for direct finish clients. Lots of the issues we do are usually not consumer-facing. They’re utilized by establishments, laboratories, and programmes moderately than particular person customers.

Venkatesh Kannaiah: How do you choose these issues?

Raghu Dharmaraju: So, usually, our strategy is that this: the federal government already runs large-scale programmes aimed toward societal impression. We construct improvements into these large-scale, already-scaled methods to enhance their effectiveness.

That’s the common considering. What’s the current system, and what’s the AI product that may assist make that system work higher?

We additionally ask whether or not such know-how is strategically vital for India. So we have to construct our personal capabilities.

Venkatesh Kannaiah: Your ideas on the AI and robotics ecosystem in India?

Raghu Dharmaraju: We’ve got the uncooked expertise. What we frequently lack are the amenities and the substantive investments wanted to do that work. And I feel that’s the place now we have began to play a reasonably important and considerably distinctive function.

While you come right here, you’ve got entry to prototyping amenities, experience, funding, and expertise — multi functional place. Many of the incubation ecosystem in India is comparatively light-touch and doesn’t present this sort of built-in help.

I additionally suppose the venture-building strategy is basically lacking. Usually, the mannequin assumes {that a} startup already exists or involves you, and then you definately assist it. There are many incubators, however innovation and enterprise constructing is a unique strategy altogether.

Venkatesh Kannaiah: What’s the largest problem you are attempting to unravel?

Raghu Dharmaraju: The most important problem we are attempting to unravel is Bodily AI. That is AI that goes into robots. There’s a bodily system, however it’s the intelligence that sits on prime of it. Humanoids are a superb instance. Bodily AI is actually what we need to go after.

To make it easy, once I ask you to select one thing up, you robotically form your hand in a sure means and choose it up. If it’s a tender mango or a strawberry, you’d watch out to not crush it. However how do you educate a robotic to do that?

The older means of doing bodily AI was to offer very exact directions. For a set, predictable activity, that works. However the second there may be even a small quantity of unpredictability, robots are inclined to fail.

So how do you educate this? That’s the place bodily AI is available in. Like all AI, it depends on coaching. Picture recognition, for instance, works by exhibiting 1000’s of photographs of cats and canines and labelling them. Over time, the system learns to inform the distinction.

Equally, how do you educate a robotic to clean a vessel or fold a chunk of material? These are extraordinarily onerous duties. You can’t realistically give exact directions for each scenario. As a substitute, it’s important to construct bodily AI, the place the intelligence emerges from the mix of software program, {hardware}, and studying.

Venkatesh Kannaiah: So is the problem primarily about ingesting huge quantities of knowledge?

Raghu Dharmaraju: It’s partly that, however not solely that. There may be innovation in coaching, within the mannequin architectures, within the computational energy required, and within the quantity and high quality of knowledge. All of those points come collectively. That’s the reason bodily AI is such a giant problem.

I imagine bodily AI is the subsequent frontier. And it isn’t nearly robots. There’s a broader idea referred to as Business 5.0. We’ve got heard of Business 4.0. Business 5.0 is about extending human capabilities by infusing duties with AI, in a human-centred means. Superior robotic methods that embody bodily AI are a core a part of Business 5.0.

Venkatesh Kannaiah: Which of your startups are working within the Bodily AI house?

Raghu Dharmaraju: Kinesthetiq is one instance. There may be Strider Robotics, which builds quadruped robots/robotic mules that may go into harmful or high-altitude environments. This might be mining, oil and fuel, or defence — conditions the place you would like to not ship individuals into hazardous circumstances. These quadruped robots can carry sensing gear, conduct surveillance or inspection, and return safely. These quadruped robots are presently in pilot levels.

The truth is, bodily AI exhibits up throughout a lot of our startups. Anyplace there may be bodily motion, robotics and AI are concerned. Drone startups, for instance, are additionally deeply rooted in bodily AI.

There may be additionally a really early-stage firm referred to as Vishwasis Aerospace. Professor Radhakant Pari, who was concerned in creating steering, navigation, and management mechanisms for Chandrayaan-3, is a co-founder. He’s now engaged on bringing what he calls physics-informed AI to drones.

The problem they’re attempting to unravel is: how do you land a drone in windy circumstances, on a transferring automobile or a ship that’s pitching on waves? That requires extraordinarily exact bodily intelligence to return collectively.

Venkatesh Kannaiah: Is it that many developed economies have already got these applied sciences and should not need to share them?

Raghu Dharmaraju: It’s extra advanced than that. In some circumstances, we do have entry to open-source fashions and analysis outputs as a result of a lot of this has emerged from world analysis ecosystems. There may be additionally a rising realisation amongst some gamers that collaboration can create extra worth total.