Humanoid robots step towards industry

2 May 2024  

A humanoid robot from Figure AI moving totes

In recent years, millions of YouTube viewers have marvelled at the capabilities of the humanoid robots being developed by Boston Dynamics. We have watched in awe at their ability to perform acrobatics, dance and carry loads. Their capabilities have improved over time, but they have been regarded largely as technological curiosities – not practical machines that could be deployed to perform useful tasks in factories and warehouses.

But a flurry of developments in recent months have revealed that humanoid robots are starting to make their way into real factory and warehouse environments. Car-makers, in particular, are showing great interest in introducing humanoids to their shopfloors. BMW, Mercedes and the Chinese automotive manufacturer Nio are all exploring the possibility of deploying these machines in their plants.

A year ago, economists at Goldman Sachs Research released a report predicting that the global market for humanoid robots would amount to $6bn as soon as 2035. Now, a year later, they have released another report, this time predicting that the market will be worth $38bn by 2035 – a six-fold increase – with 1.4 million humanoid robots likely to be shipped a year by then – a four-fold increase on last year’s forecast. The researchers are also predicting that the sector to become profitable much sooner than they were expecting a year ago.

The researchers offer several reasons for their drastically revised figures, including a 40% reduction in the cost of materials, the accelerating progress in AI (artificial intelligence), and investment in the humanoid sector expanding much faster than they had anticipated.

“AI progress surprised us the most,” the team says in the new report, referring to advances such as robotic large language models (LLMs). They point out that there’s been significant progress in end-to-end AI, through which models can train themselves, removing the need for human engineers to code everything by hand. That’s speeding up robot development, allowing these devices to handle more tasks and to adapt to new situations (such as working outside of factories) more quickly.

Perhaps the most dramatic example of the money now being poured into developing humanoid robots comes from a two-year-old Californian start-up called Figure which has yet to produce a commercial product, but has recently attracted $675m of funding from a group that includes Microsoft, nVidia, Intel and Amazon’s boss, Jeff Bezos. Figure is also working with the ChatGPT developer OpenAI to integrate AI capabilities into its humanoids.

Cheaper components

Goldman Sachs says there are signs that key robot components – ranging from high-precision gears to actuators – could also cost much less than previously expected, leading to faster commercialisation. The manufacturing cost of humanoid robots has dropped – from a range that ran between around $50,000 (for lower-end models) to $250,000 (for state-of-the art versions) last year, to $30,000–$150,000 now. A year ago, the analysts had expected costs to fall by 15-20% a year when, in reality, these costs have fallen by around 40% over the past year alone.

“We expect further cost reduction in the coming years,” the report states. As well as cheaper components, there are now more supply chain options, and designs and manufacturing techniques have improved, it adds. This could, in turn, speed up the timeline to the deployment of humanoid robots on factory applications by about a year, and in consumer applications by two to four years, compared to last year’s estimates.

The researchers’ base case is now for more than 250,000 humanoid robots to be shipped in 2030 – almost all of which would be for use in industry. Sales of consumer robots will then ramp up quickly over the coming decade, with more than a million of them being shipped annually in just over a decade. Medical uses, and commercial applications such as folding laundry, will also benefit.

Geographically, no one country or region appears to be dominating this emerging sector. While Western companies are likely have the most sophisticated AI software models, Asia will probably become the manufacturing hub for humanoid components, because of its wide supply chain base and lower manufacturing costs.

Another reason that Goldman Sachs Research is more optimistic about the prospects for the sector is that more players are pouring resources into developing and manufacturing humanoid robots. The Chinese government started a fund to drive robot R&D, while component-makers are recruiting staff and devoting capital and human resources to their activities.

According to Goldman Sachs, most of the hardware needed to create humanoid robots is already available, or close to maturity. Components such as cameras, motors, force sensors, precision gears and batteries are mostly ready for commercial use.

Hurdles and bottlenecks

But there are still hurdles to overcome. Some components, for example, need precision grinding machines that are limited in number, making it difficult to ramp up production. Costs for some components are still high because of restricted production capacities or long manufacturing cycle times.

And there are still significant bottlenecks in the development of AI and software for robot manipulation, such as grasping objects, and human interaction – taking spoken commands without training, for example. Goldman Sachs is optimistic that these remaining barriers will eventually be overcome, leading to mass-produced, general-purpose humanoid robots, although it concedes that the viability of such machines has yet to be proven.

The bank expects significant demand for humanoid robots in structured environments such as manufacturing using technologies that are available today. This could include applications such as assembling electric vehicles and sorting components. About 70% of manufacturing in China is already done by machinery and automation, while only 20% is handled by manual labour, and 10% done by managers. Because humanoids are more flexible and capable of adapting to complex environments, the analysts believe that bipedal robots can expand the industrial automation market.

Humanoids are particularly appealing for tasks that are dangerous, dirty, and dull, the report states. The Goldman Sachs researchers envisage potential demand for the mobile robots in areas such as mining, disaster rescue, nuclear reactor maintenance, and chemicals manufacturing. They point out that users in these sectors may be willing to pay a higher price for robots that can do dangerous jobs that people are reluctant to perform. In addition, robots could also augment labour in sectors experiencing shortages of human workers.

Assuming a labour substitution rate of 5-15% for car manufacturing, as well as for dangerous jobs such as disaster rescue and working in nuclear power plants, the reports predicts an the annual global demand for 1.1–3.5 million humanoid robots. The analysts suggest the best investment opportunities at present could lie in the component manufacturers that are vital to the supply chain.

In a “blue-sky” scenario, in which innovation unfolds rapidly and demand soars, Goldman Sachs Research envisage humanoid robots becoming the next “must-have” device, similar to smartphones or EVs. Such robots would become vital for manufacturing and dangerous work, but they would also help with caring for the elderly and fill in for labour shortages in factories.

The report identifies and compares nine established non-Chinese humanoid robot developers – mainly the US – including Tesla, Boston Dynamics and Kawasaki. Their robots can move at speeds of up to 3m/s, and operate with up to 74 degrees of freedom. But there are also at least ten humanoid robot developers at work in China, and there may be others which have not been made public yet.

According to Goldman Sachs’ revised forecast, humanoid robots could reach two-year back backs in some factory applications as soon as 2025-2026 (up to two years sooner than it was suggesting a year ago). In consumer applications, two-year paybacks could be a reality by 2028-2031 – compared to the 2030-2035 timeframe it was quoting last year.