At the beginning of this century, when Honda presented ASIMO to the world, it was a watershed moment in the history of humanoid robots. Soon, ASIMO started appearing in celebrity photoshoots, and its clumsy dance moves were getting itched in people’s memory. 

A decade later, Boston Dynamics debuted their famous canine-like robot Spot, which immediately became a sensation with its viral YouTube videos. Fast-forward to the present day, we just saw an Optimus robot mingling with humans and serving drinks at the last Tesla’s RoboTaxi launch party. These three distinct events depict how this technology evolved and how we are at pivotal moments to see the transformation of human labor participation.

Humanoid robots have occupied our imaginations for a long time. Starting with Isacc Asimov’s ‘I, Robot’ to R2D2 and C3PO of Star Wars or let it be the Robot Cop roaming the streets, they have been a staple in popular science fiction. 

Technological advances in computer vision and sensing, innovations in battery technologies, and material sciences seem to have propelled us to the modern day when these things started to pop up in real life outside the book’s pages. 

Countries with low birth rates and an aging population will soon face a shortage of human labor, and companies working on humanoid robots are aiming to address this problem. The pace of developments and the market opportunity are so vast that Goldman Sachs has recently revised its prediction and is forecasting a 38 billion USD market by 2035. 

What is driving the innovation

One cannot help but wonder why startups and large corporations are suddenly investing in humanoid robots and why innovation is accelerating so rapidly. The answer is multifaceted. Machine learning technologies centered around computer vision, sensing, and perception have made giant leaps in the last 15 years. 

Development in GPU technology, fundamental discoveries around machine learning techniques, and edge computing have enabled on-the-go real-time object recognition. We see these advances in autonomous driving technologies. Hauling an autonomous Uber ride in San Francisco or getting food delivered by a delivery robot on US college campuses has become common. However, developing humanoids requires solving a few additional hardware challenges, too. 

Two key problems must be solved for a robot to mimic human movements. One is to balance itself during motion. Unlike autonomous vehicles, which have at least 3 points of contact on the ground, a human-like movement can only have at most 2 points of contact. Companies like Boston Dynamics and Agility Robotics have developed actuators and other components that make a human-like gait a reality. 

Humans also have fine motor skills and can grab, manipulate, and rotate objects. Advancements in sensors, artificial intelligence, and materials enable robotic hands to transition from basic industrial tasks to more complex operations such as delicate surgical procedures, intricate automotive assembly, and personal care. 

The price of high-precision gears to actuators seems to be dropping to a point where humanoid robots can become affordable. According to a report by Goldman Sachs, the manufacturing cost of humanoid robots has decreased by 40%, bringing down the price range from $50,000 -$250,000 to $30,000-$150,000. Although we are far from fully functioning human replacements, recent advances have made venture capitalists and large companies increase their investments. 

In that spirit, Accenture has recently invested in Sanctuary AI, a developer of humanoid general-purpose robots powered by AI that can perform a wide variety of work tasks quickly, safely, and effectively. According to Accenture, this strategic investment will play a key role in the future, when robots will augment human workers to perform manufacturing tasks and reinvent how work gets done efficiently. 

The players

The last three years’ holiday news cycle has been dominated by extremely bad working environments at Amazon and other similar retailers’ warehouses. Long working hours and fewer breaks to keep up with the ever-increasing order fulfillment target make these jobs unsafe and have tainted the public image of these companies. 

So, when Amazon announced in April 2024 that it would invest 700 million euros in AI and robotics to improve its European fulfillment centers, we knew it was coming. Amazon currently has 750,000 robots deployed in its warehouses, which makes it the largest of its kind. 

Contrary to popular perception, Amazon robots are not eliminating human jobs; they augment humans to increase productivity and reduce unsafe working conditions. Along with its in-house innovation, Amazon continues to invest in startups like Agility AI to reap the benefits. Amazon’s Robin robotic arm has recently surpassed a cumulative total of 3 billion picks. Given the diverse nature of the objects it had to pick, this symbolizes how far the dexterity of these robots has come along. 

While Amazon is investing in solving the grip motor skill replication problem, it has shied away from addressing the gait problem. This arena is for veteran companies like Boston Dynamics and new players like Figure AI. We have all seen the demos of Boston Dynamics’ Atlas, which does a marvelous job of keeping the balance on two feet when navigating various terrains or getting perturbed by external agents. 

In January 2024, Figure AI announced a partnership with BMW to test out its humanoid robot, Figure 02, in their South Carolina plant. New human-scale hands, six RGB cameras, and perception AI models trained with synthetic data generated in Nvidia’s Isaac Sim enable Figure 02 to perform high-precision pick-and-place tasks required for smart manufacturing applications. 

These robots have been found installing sheet metal onto car chassis, an ergonomically awkward and tiring task. Following their initial announcement, Figure AI also raised $600 million in its Series B, with a significant investment from Nvidia. Their collaboration with Microsoft and OpenAI also makes them a key contender against Boston Dynamics and Tesla.

Outside of these hardware companies, the software ecosystem for robots has evolved significantly in the last five years. Traditionally, the perception, thinking, and movement components were modular, with rigid interfaces interconnecting them. These interfaces often made the robots slow to react to environmental changes, resulting in less fluid movements. 

Nvidia is addressing this by building general-purpose foundation models for robots that utilize multimodal inputs, such as visual, audio, and sensor data, integrating all aspects of perception, reasoning, and control into a unified model. These models use historical interactions and multimodal data to enable robots to make complex decisions and perform sophisticated tasks autonomously. 

The company has also launched a specialized Humanoid Robot Developer Program, offering early access to NVIDIA’s AI foundation models, robot learning tools, accelerated libraries, simulation frameworks, Jetson™ Thor computing platform, and robotics workflow orchestration services for humanoid software, hardware, or robot manufacturers.

Impact and opportunities

In countries with lower labor costs, the math still favors the human labor force. However, as the production costs of these robots continue to come down and their effectiveness and efficiency keep increasing, we will soon reach the pivot point where investing in these technologies will become more lucrative than maintaining a human labor force. 

AI-powered robots will have the right skill set, reducing on-the-job training, fewer errors, and longer productive hours. This raises the critical question of what to do with the labor pool. 

Amazon’s approach to installing more robots in their fulfillment centers has been seen as a corporate giant’s ulterior motive to replace human dependency. However, the truth is that the lower birth rate and an aging population are reducing the labor participation rate in the US and EU, and to maintain the same amount of productivity, this is a much-needed investment. 

Though it is not the same case yet in the developing world, a robot-centric universe will soon become a reality there, too. Strategic investment in upskilling the labor force can prepare nations to navigate this future successfully. New job categories will emerge for developing and maintaining these robot workers. Both public and private investment in education centered around this new ecosystem is needed to capture these new opportunities. 

The augmentation of humanoid robots with a human labor pool is bound to extend human capability and productivity. Although we are far away from the future of entirely replacing the human labor pool, the pace of innovation in this domain is ever-increasing. It is high time to let both private and public investment follow and lay the foundation for the upcoming transition.  

Sameeul Bashir is a computational scientist currently working at the National Institutes of Health in the United States.