The coming of age of India’s AI-powered humanoids

She is no Sophia, the famed humanoid robot from Hanson Robotics that is now a citizen of Saudi Arabia, in terms of looks. And she uses wheels to move around, instead of legs.

But then, Anushka, which stands for ‘Artificially Networked Unit for Smart Human Knowledge Assistance’, was “developed for less than ₹2.5 lakh with some components sourced from the dump yard of our mechanical engineering department”, Piyush Khanna, a robotics student at KIET, told Mint.

In comparison, a polished and sophisticated robot like Sophia would cost anywhere upwards of ₹25 lakh.

“Anushka is still being tested. A silicone mask and legs would make her look more realistic,” acknowledged Khanna, but added that it does imply that “she is not smart like Sophia.”

Anushka, for instance, uses computer vision to recognize faces, roll her eyes up, down, and sideways, follow people as they walk, and even make automated phone calls to specified people in case her batteries need recharging, or if someone threatens her.

The robot can perform over 50 hand gestures, 30 eye gestures, along with jaw and neck movements. It moves around on a three-dimensional (3D)-printed base with wheels, and uses servo motors for precise motion control, and “realistic” lip movements.

Anushka has an i7 processor, 16 GB RAM (memory), and uses natural language processing (NLP) to understand and respond to questions. She can understand 60 languages but currently responds only in English.

Her other features include home automation to control lights and the ability to make real-time decisions using reinforcement learning like humans do.

“Anushka also has a beating heart and uses the Internet of Things (IoT) to make the calls,” added Shubham Shukla, who heads the Centre of Excellence for Robotics and Mechatronics, and mentors Khanna along with Vibhav Sachan, head of electronics and communications union department at KIET, and Himanshu Chaudhary.

Growing tribe

Anuskha is just one of the many humanoid robots that Indian organisations and companies have built, or have been building, over the past few years. In 2015, a two-foot-tall humanoid robot called Manav created a stir when it made its debut at Techfest 2014-15 conducted by IIT Bombay. It had 21 sensors, two cameras in its eye sockets and two mikes on either side of its head and was India’s first 3D-printed humanoid robot.

In 2021, Dinesh Patel, a computer science teacher at Kendriya Vidyala in IIT Bombay, developed a humanoid robot called ‘Shalu’ using waste material including plastic, cardboard, wood and aluminium. The prototype robot could recognize people, shake hands with them, and answer their questions.

Other humanoid robots include HDFC’s IRA robot, India’s airport assistance robot called RADA, H-Bots Robotics police robot dubbed Robocop, and Rashmi, a robot that could speak Hindi, Bhojpuri, Marathi and English.

Many projects including IRA, though, did not go beyond the prototype stage. But the efforts continue.

Take the case of Ahmedabad-based Kody Technolab, which introduced Skanda, its AI-powered humanoid robot at the Vibrant Gujarat 2024 Summit. And in May, the startup’s autonomous surveillance robot Athena, and serving robot Dasher, held fort during the Tuneland Music Festival in GIFT City, Gandhinagar.

Likewise, Balaji Viswanathan, CEO of Bengaluru-based Mitra RobotIndia Pvt. Ltd (the US entity is called Invento Research Inc., and the startup does business there as Invento Robotics) is known for its Mitra and Mitri humanoid robots that are now being positioned as “senior care companion robots.”

“We currently have about 50 robots operating in the U.S., with additional pilot projects under way in Singapore and Australia. Our primary focus is on addressing the needs of senior care, a market that presents unique challenges due to its vulnerable population,” he said over a call from his California office.

“Ensuring safety and efficacy is our main priority, so we’re not aiming for immediate large-scale deployment but rather focusing on solving day-to-day issues effectively.”

Mitra’s business model includes renting these robots to generate revenue, alongside the core focus on healthcare solutions. This rental service is a secondary aspect of the operations, providing robots for short-term use, including specifically tailored applications for Fortune 500 companies, like monitoring factory equipment for compliance and safety.

Currently, 12 robots are in circulation for rental purposes, typically 2-3 days across various use cases.

Wheels vs legs

India’s humanoid robots will only get better with time. Anushka’s successor Vinayak will walk on legs instead of moving on wheels and have enhanced social interaction, according to Shukla. The goal is to position these robots in industries such as entertainment, medical assistance, and even Bollywood, where realistic robotic characters could replace traditional animations.

Viswanathan is working on making Mitra more intelligent and has tied up with a team led by Ramana Vinjamuri, an expert in brain-computer interfaces (BCIs) at the University of Maryland. The goal is to help individuals with paralysis overcome daily challenges using robots like Mitra.

The system involves a non-invasive BCI headset that detects muscular impulses when a person intends to perform a task, such as grabbing a bottle of water. Although their muscles can’t complete the action, the BCI reads the intent and transmits the signal to the robot, which then performs the task, he explained. This approach avoids some limitations of exoskeletons such as wearability issues and is particularly useful in settings like senior care.

While the University of Maryland is handling the BCI technology, Viswanath is focusing on the robotics aspect. Future developments may include a more integrated, user-friendly headset that functions as a peripheral to the robot system. The project, though, is still in the research phase and not commercially available.

The global humanoid robot market, according to The Brainy Insights, a markets research company, was valued at $1.38 billion in 2023 and is forecast to touch $138.09 billion by 2032. They find use in the healthcare sector for caregiving, improving product launches, and even for dangerous tasks such as space travel, underwater investigation and lifesaving.

Stanford Robotics Lab designed Ocean One, a bimanual underwater humanoid robot, to study coral reefs. And in February, the Indian government announced that a woman robot astronaut ‘Vyommitra (space friend in Sanskrit)’ will fly into space this year. Vyommitra is equipped to monitor module parameters, issue alerts, and execute life-support operations.

What is also helping is that companies such as SoftBank Robotics in Japan and PAL Robotics in Spain have invested substantially in humanoid robots and integrated them into their business strategies, particularly targeting the service industry, including healthcare, hospitality, and retail, notes Grand View Research.

Last March, OpenAI invested $23.5 million in 1X’s Bipedal Humanoid Robot NEO. This March, Nvidia announced Project GR00T, a foundational model aimed at humanoid robots to advance its initiatives in enhancing robotics and making significant progress in embodied AI.

Overcoming limitations

Wheel-drive robots are easier to design and program than legged ones and are typically operated as entertainment sources at science events, theme parks, and amusement parks. They are forecast to become more prevalent in military and defence applications. Legs provide humanoids more flexibility but are far more expensive.

Also, while companies such as Boston Dynamics have demonstrated impressive robot capabilities, their machines typically have limited battery life. And current industrial robots often operate on fixed axes and are typically wired, making them unsuitable for versatile, battery-operated humanoid applications. Privacy and security, too, will pose major hurdles as these sophisticated robots grow more networked and become prone to cyberattacks.

Further, developing humanoid robots presents significant challenges, particularly in replicating human muscle efficiency, according to Viswanathan. Unlike brain functions, which can be outsourced to cloud computing, muscle tasks like lifting and moving items must occur on-site, requiring immense energy that current technologies struggle to supply efficiently.

While human muscles can perform thousands of actions effortlessly, robots need a lot of energy, which existing battery technologies like lithium-ion cannot deliver quickly enough without overheating. Capacitors offer potential because they can release energy instantly, but integrating this into humanoid robots remains complex, he added.

Viswanathan’s company is focusing on memory functions and patient-specific customization rather than solving the muscle challenge this year, leveraging BCI and LLMs to enhance robot-patient interactions and predictive abilities.

There’s another problem.

“In the US, our robots typically cost about $500 per month, which is affordable by local standards. However, in India, the equivalent cost (about ₹42,000) makes them less competitive against cheaper human labour, posing a challenge for broader adoption. We’re currently running pilots in India, including collaborations with Apollo and Fortis hospitals, exploring customised robotic solutions in healthcare settings,” Viswanathan said.

He added that medical assistant robots can particularly help in institutional settings like senior living facilities where one nurse may have to manage the diverse needs of many residents. The robots can automate tasks like monitoring patient safety, alerting staff to emergencies, and handling basic requests, allowing nurses to focus more on human interaction and care, explained Viswanathan.

Shukla and Khanna corroborated that developing the Anushka humanoid robot presented several challenges. Integrating different modules, such as computer vision and NLP, led to anomalies.

For instance, Anushka could hear and respond to speech, but the speech-to-text and text-to-speech modules conflicted, causing the robot to listen to and respond to its own output. A master programme was needed to manage the timing of these processes, turning the microphone on and off as required, Shukla explained.

Anushka’s learning capabilities also posed challenges. Initially, she was designed to learn through reinforcement from user interactions. However, due to the varied dialects and potentially inappropriate language of users, her learning had to be temporarily halted and replaced with custom training to ensure appropriate responses.

Last, but not least, there were challenges in designing realistic facial movements, which led to a collaboration with a Bollywood-associated silicone face mask maker to develop more lifelike and movable facial features.

“If some industry or some government body is ready to fund us, I think we can design robots that are better than Sophia. There are features in Anushka which are more reliable and more profound than Sophia – the only thing we are lacking is the finishing, which we will be able to overcome once we have proper funding,” Shukla said.

While this may sound like a tall claim, Shukla does make a valid point for the whole industry.