What does it take to move robot dogs from novelty to real-world deployment? Aditya Pratap Singh Rajawat from xTerra Robotics explains to EFY’s Nidhi Agarwal how these systems are being built for industry and defence applications.
Q. What does your company do?
A. xTerra Robotics is a deep-tech startup building advanced legged robots, four-legged robot dogs and humanoids. We develop the necessary subsystems in-house, from actuators and electronics to the body and control systems, enabling robots to operate reliably in complex, real-world environments.
Q. What inspired you to start your company, and is there a story behind its name?
A. We started the company out of the research that we were doing at the Indian Institute of Technology (IIT) Kanpur on legged robots, beginning in 2019 with Prof. Shakti Gupta and Prof. Mangal Kothari and working closely with the Defence Research and Development Organisation (DRDO). Over the years, we built early lab prototypes and a passionate team. After some of us graduated and while we worked in industry, we stayed connected to the team. By 2023, we saw a clear gap. There were almost no deployable robots in India, while countries like the US and China were moving much faster. We realised we had the chance to build robots from the ground up in India and serve growing global markets. With our prototype and team ready, we started xTerra Robotics in March 2023, and within six months, we delivered our first robot to a customer.
As for the name, ‘xTerra’ comes from ‘x’ and ‘Terra.’ ‘x’ represents any variable, and ‘Terra’ means terrain. It reflects our goal to build robots that can operate on any terrain.
Q. How many founders started the company?
A. The company has five co-founders. Most of them, including me, are from IIT Kanpur. One is a professor, and the other four worked with the professor in the robotics lab. Some co-founders have PhDs, while I have completed my bachelor’s and master’s degrees.
Q. How many employees does xTerra Robotics have, and how is the team structured?
A. We currently have 15 people, including the five co-founders. Eight members work in research and development (R&D), and two handle production and operations. Most individuals handle multiple roles contributing to both new product development and the ongoing manufacturing of Actuators, Cobots, and Svan M2. We also work closely with IIT Kanpur’s Mobile Robotics Lab, which comprises 10 members who support our R&D efforts.
Q. Can you discuss a few of your products?
A. One of our key products is Svan M2, India’s first commercial quadruped robot, or robot dog. It is designed for research and industrial users across agriculture, factory automation, security and defence. It weighs about 11kg and can carry a 5kg payload for nearly an hour at a speed of nearly a m/sec. With onboard sensors/robotic arm, it can detect gas leaks, spot anomalies on factory floors, and support border security and surveillance, helping replace or support humans in dull, dirty, and dangerous tasks.
Q. You say Svan M2 is India’s first commercial quadruped robot, but other robots are already sold on distributor sites. How is this claim correct?
A. We claim this because we started the work as a research project at IIT Kanpur in 2019, at a time when no such robot was available to buy, either in India or globally. Even global products like Boston Dynamics’ Spot became commercially available only in 2020 and were not sold in India due to IP restrictions.
We founded xTerra Robotics in March 2023, got our first order within three months, and delivered the first Indian-made commercial quadruped robot to IIT Palakkad in January 2024. At that time, no other Indian company had commercially sold and delivered such a robot. Today, a few Chinese products are sold in India through resellers, and some Indian players have emerged, but they have not yet commercialised their products. That is why we can say Svan M2 is India’s first commercial quadruped robot.
Q. What is the Cobot C1?
A. The Cobot C1 is a collaborative robotic arm that can be mounted on the Svan M2 robot. It is designed for tasks like inspecting pipes, operating switches, and moving objects. We designed and developed the lightweight arm such that its size and reach fit the robot perfectly. It can access the ground from all directions without overloading the robot. The arm also carries cameras for inspection and grippers for pick-and-place tasks. The term ‘cobot’ comes from ‘collaborative robot,’ which differs from traditional industrial robots that are usually caged and unsafe to interact with. A cobot, like C1, can be safely held, guided, and programmed to replicate tasks, making it a subclass of robots designed to work alongside humans.
Q. Can you explain what is unique about your robot hardware in the overall design?
A. The key part of our hardware is the actuator that drives each joint. Each actuator includes a motor, driver, gearbox, and housing, all packaged into a compact case. Our goal is to make it as small and lightweight as possible while still delivering maximum torque. These actuators are compliant and back-drivable, meaning that if someone applies an external force, the joint moves naturally rather than holding its position. This makes the robot safer, more responsive, agile, and fast, both when reacting to obstacles and when moving at higher speeds to perform tasks. A major part of our innovation is the compact gearbox design and the way everything is packaged, which we have patented and made proprietary. We are continuously improving this design to increase torque output while reducing weight.
At the system level, this actuator design allows us to build a complete robot body that is lightweight yet capable of carrying significant payloads. The challenge is to keep the self-weight low while ensuring enough torque to handle loads; the robot would struggle under its own weight. For example, our Svan M2 robot has a very high payload-to-weight ratio. This performance is unmatched in its category, both across India and globally. Despite being the lightest robot in its class, it can still deliver enough torque to carry heavy loads and perform tasks efficiently.
Q. What hardware challenges did you face while developing your commercially available legged robot, and how did you resolve them?
A. Actually, first we developed the Svan M1, whose body was made of carbon fibre; however, its actuators were made of high-grade aluminium. Next, building on lessons learned from Svan M1, we systematically migrated to an aluminium body and torso and upgraded the actuator to achieve a higher torque-to-weight ratio in Svan M2. While in theory one can reach an optimised design, manufacturing it precisely could be very challenging. Specifically, minimising backlash in gears and the play between mating parts. Additional issues include impulsive loads on the robot’s legs and torso, and heat generation. Impulses lead to vibrations, looseness of bolts/screws, fatigue failure, and heating, which leads to thermal expansion and degradation of the motor driver’s performance. We need to guard against these failure modes.
To overcome this, we partnered with an aerospace-grade manufacturer capable of delivering higher-quality components after multiple iterations. Through these iterations, both our team and the manufacturer learned. Our designers learned how to relax certain constraints/tolerances without compromising function, and the manufacturer learned how to optimise production within their limitations. After building five versions of the robot with incremental improvements, we achieved repeatable, reliable hardware that works indoors and outdoors and can handle rigorous testing to its limits.
To be honest, we keep improving the hardware from batch to batch as we learn from previous batches! Unlike wheeled robots, legged robots are challenging to manufacture.
Q. What challenges did you face in developing the software for the robot?
A. We faced several challenges across both software and AI/ML implementation. From the first line of code, we had to ensure the system could operate in real time, often running at 500Hz, which is 500 cycles every second. At this speed, failures can occur either because the system cannot process everything fast enough or because of communication delays. Optimising for low-power compute added complexity, since we could not just use the most powerful computers without draining the robot’s battery.
Real-time responsiveness was a major challenge because this robot does not move freely. It is continuously hitting the ground. Earlier versions made the robot’s motion jittery or caused it to topple, so we optimised both the software and communication layers to make it truly reactive. On the AI front, we had to process continuous camera feeds and sensor data efficiently. Without optimisation, we might get one frame every two seconds, making obstacle avoidance and environment mapping impossible. Cleaning sensor noise, improving processing rates, and managing computational resources were all part of an iterative process to make the system responsive, reactive, and reliable.
Q. What sensors are used on your robots, and how does sensor fusion work in your system?
A. Our robots use several sensors to track motion and orientation. Each joint has an encoder that continuously measures position and velocity. At the centre of a robot’s torso, an inertial measurement unit (IMU) measures accelerations and angular rates to determine the robot’s orientation, helping it maintain balance. Sensor fusion combines data from all encoders and the IMU to create a unified understanding of the robot’s state, accounting for noise or differences in sensor rates. This allows the robot to know its leg positions, body orientation, and how it interacts with the terrain.
In addition to motion sensors, the robot uses cameras and LiDAR to perceive the environment. Cameras provide continuous images and depth information, while LiDAR creates 3D maps with laser points. Fusing these sensors helps the robot navigate autonomously, avoid obstacles, and replan paths in real time. We also add application-specific sensors, such as thermal or pan-tilt-zoom cameras, for inspection and surveillance. These capture thermal images or detailed visuals of factory equipment, which are analysed by AI models while the robot moves across the floor.
Q. How do you ensure robots are safe to operate around humans in shared spaces?
A. We ensure safety through both hardware and software design. We use compliant actuators, so the joints yield when touched instead of applying force. If a person interferes, our robot senses the disturbance and changes direction or moves away. Unlike industrial robots that follow fixed motions and can cause injury, our robots react to contact, maintain balance, and avoid obstacles. At a higher level, our AI plans movements to prevent collisions in advance, making operations around humans safer.
Q. How do you ensure that your lightweight robots have enough power to operate for long periods? How does the battery system work?
A. For our current Svan M2 robot, we use a bought-out, replaceable battery pack. It runs for about 1 hour, then can be swapped out for another battery to continue operating. With our new product, we are assembling custom battery packs using commercially available cells. These packs are shaped to fit the robot and can allow operation for over three hours. We are also exploring wireless charging so the robot can dock and charge itself. The goal is to eventually reach an eight-hour runtime, matching a typical human-shift in a factory setting. We leverage the best available battery technology from the market and integrate it to meet our specific endurance and weight requirements, keeping the robot lightweight while extending operational time.
Q. How are you using AI and ML in your robots?
A. We are using AI in two main ways in our robots. First, AI helps robots move by training them to adapt to slippery surfaces, uneven ground, and other disturbances. Second, AI helps them observe and react by analysing video or images in real time. The robot processes this data on board using its own processor and GPU, understands the situation, and takes actions similar to those a human would.
Q. Who are your target customers, and what problems do you solve for them?
A. Our target customers are asset-heavy industries and the military/para-military/civilian law enforcement forces. In the industry, we focus on power, oil and gas, automotive, and pharmaceutical sectors, where many assets need monitoring. We provide mobile robots that collect sensor data and also act on it, such as operating switches or closing valves, reducing IoT sensor costs and enabling real-time response. These robots operate safely around people and in complex spaces such as factory floors, staircases, and public areas.
For military/para-military/civilian law enforcement forces, our focus is on intelligence, surveillance, target acquisition, reconnaissance, and logistics. Our robots can carry payloads ranging from 5kg to 15kg today, and we are developing versions capable of carrying payloads over 50kg to transport supplies, sensors, and weapons/equipment across rough terrain. While these forces are not yet our customers, they are our key targets. We are actively working with defence labs and defence public sector undertakings (PSUs) to develop these capabilities.
Q. Who are the early adopters of commercial robots in India, and what has been their experience using them?
A. As far as I know, the early adopters of commercial robots in India are the Indian Army anda pharmaceutical industry. I understand that the Indian Army purchased 100 robot dogs from a US company, Ghost Robotics, for ₹2.85 billion. These robots carry 15kg and have limited operational flexibility. While there is interest in deploying them in regions like Jaisalmer or Ladakh, challenges remain in fully utilising them due to payload limits, operational constraints, and high costs. Some investments have been made to use these robots, but many attempts have failed because the available robots do not fully meet operational needs.
In the industrial sector, a pharmaceutical company bought a Chinese robot to automate audit inspections, expecting it to collect data, detect spills, and monitor factory floors. However, they could not get support from the manufacturer or resellers, making the robot unusable for their needs. They approached us, and we adapted our expertise to make both their robot and our hardware perform useful work. A one-month pilot completed in October successfully automated audit inspections, highlighting a gap in the market: early customers often need local expertise to be operational. Because of this, a company prefers vendors that provide full hardware and software control to enable scalable deployment across multiple factories.
Q. How do you view the competitive landscape in quadruped robotics, and how do you differentiate your technology from other players?
A. Globally, the quadruped robotics field is dominated by a few major players: Boston Dynamics and several Chinese companies. Boston Dynamics has commercialised its technology over the years, supplying over 3000 robots worldwide, but its legacy hardware and software limit its ability to adapt rapidly. Chinese players focus on affordable robots, but these often remain “toys” without full industrial usability.
We develop a full-stack, industry-ready solution that is compact, affordable, and highly adaptable. Our robots operate in extreme conditions, run continuously for hours, and can be customised for specific industry needs at a fraction of competitors’ costs. With the world’s best payload-to-weight ratio, combining in-house design, customer-driven customisation, and rapid adaptability to deliver unmatched industrial performance.
Q. Do you sell actuators in addition to robots, and do you also provide client-specific robotic solutions? How does this contribute to your revenue?
A. Yes, we do supply actuators. Earlier versions were built for labs and organisations creating custom robotic arms, which can use our actuators to build their systems. xTerra-made actuators are also available on marketplaces like Robu, or directly through us or our resellers. They are often used when smaller robotic arms are needed for desk-level tasks or automation. Regarding revenue, the largest share comes from robots, as they are high-value items, while actuators contribute less. In addition, we develop client-specific robotic solutions tailored to specific tasks. For example, in the pharmaceutical sector, some companies needed a solution that no other supplier in India or even Chinese OEMs could provide. We collaborate with these clients to develop the complete solution, which also helps us improve our own products.
Q. How many units have you sold so far, and what was the revenue recorded?
A. So far, across all our products, including Cobot C1 and Svan M2, we have 12 different customers. Revenue to date has crossed 10 million.
Q. Where do you manufacture your robots? Do you have your own plant, or do you outsource production?
A. Most of the assembly and integration happens in Kanpur, but the manufacturing is outsourced to a few partners in the Bengaluru area. We provide them with manufacturing drawings, and after iterating with the last four to five robots, they are now well-versed in building these units. They act as our fixed manufacturing partners, and we also help them expand to make the process more production-ready and scalable. We handle integration in Kanpur.
Q. Are you receiving any funding or support for developing your robot, and if so, from where?
A. Yes, we have received support mainly through incubators. Our company is incubated at IIT Kanpur, where programs like the Social Innovation Lab provide government grants for specific projects. We also got support from the I-Hub Foundation for Cobotics at IIT Delhi. Additionally, the new Immersive Technology and Entrepreneurship Labs incubator in Chennai awarded us a grant to support product development. We have not raised any external funding outside of these incubators or government programs. Most of our operations are funded by the revenue we generate, which even exceeds the support we have received from these incubators.
Q. What challenges did you face in turning a research-focused project into a fully developed commercial product?
A. The biggest challenge was making the product reliable and functional outside the lab. In research, you can move fast, assemble pieces, and iterate to publish results. For a commercial product, everything needs to be polished, tested, and reliable for months or even years of use, which requires repeated hardware cycles and thorough testing. Certifications in India added complexity, as options for these kinds of products are limited. Achieving long-term reliability, product life of several years, and planning for maintenance contracts required a significant shift in approach and mindset from research to engineering.
The founders were pursuing PhDs and had to leave their studies to focus on product development. Unlike research, where success is measured in publications, building a product is more about engineering, designing, breaking, fixing, and certifying it. Some team members, including our Director, who was a DRDO scientist for 16 years, had prior industry experience and had delivered end-to-end products, which helped the transition. From 2019 to 2023, we moved from a proof of concept to a full product, requiring a complete rethink of our priorities, approach, and mindset.
Q. What are the main challenges your startup is facing to grow quickly, and how are you addressing them?
A. One of the biggest challenges we face when entering advanced or complex fields is talent. Along with funding, talent is a key factor for moving and building fast. Funding has been difficult, but we are about to close deals. Most of the expertise in these kinds of robots is limited, so we have developed much of it in-house. We hire and train people to deliver on these projects.
We also work with research organisations and universities, such as the Indian Institutes of Technology (IITs) and the National Institutes of Technology (NITs). By supplying them with our products, we give students early exposure, and some are trained by us. They can even reach out to us directly for guidance. This helps us when hiring from these colleges later. You see, talent is harder to find than funding, so we develop it in-house and train students at IITs and NITs. Still, there is a significant shortage of skilled talent, and we continue to address it through training and collaborations.
Q. How is your ecosystem helping with growth, and are you looking for new channel partners, distributors, or vendors?
A. Our ecosystem is growing. Currently, we work with resellers in Kanpur and Gujarat, who support us in both north and south India, and we also engage with government customers through GeM. At this point, we have not scaled enough to have multiple channel partners across India, but the drone ecosystem is expanding, and we plan to leverage it. Through these partners, we can reach technical industries, and we see a trend toward “robotics as a service,” similar to drones-as-a-service. This growing ecosystem and integration with channel partners will help us be in the right place at the right time.
Q. Are you working on any upcoming robot models or technologies that you can share?
A. We have a new robot that is not public yet, but it is planned for release around July. We’ve already built it and are testing it with some early customers, refining it to be solution-specific. It’s a similar type of robot to what we have done before, but with a much better payload-to-weight ratio. For example, the army has a robot weighing about 55-60kg that carries 15kg, while our new robot weighs close to 20kg and can carry more than 15kg. It is still in customer testing, but it should be available from July onwards.
Q. What are your plans for future growth, and where are you investing right now?
A. We are investing across the board in equipment, tools, people, and marketing, but our main focus is R&D. We are working on building the first mass-manufacturable version of our robots. Currently, we are the first to commercialise them and can produce about five robots per quarter. Over the next year, we aim to increase production to 50 robots per quarter for one product line, with additional new product lines planned. Ultimately, our goal is to set up a factory capable of producing over 1000 robots per quarter.
