The robots are here! We examine the rapid advancements in humanoid robots and their AI-powered brains. Figure 02, Atlas and Optimus as well as the exploding Chinese market. Why now? And what does all this have to do with AI? A lot; and it’s not slowing down. The societal impacts of embodied AI, the future of work and the ethical considerations as we share our homes, factories and offices with a new kind of worker.
AI Professional – Fiona Passantino, late September 2024
Go Figure
On August 6, Figure AI launched its second-generation Figure 02 with a short, flashy video showing a well-lit, sleek humanoid robot casually flexing its fingers and looking around. It was a scene straight from our childhood dreams and nightmares of our distant future.
But here it is, now. The strongest humanoid robot yet, the Figure bots feature a built-in OpenAI Large Language Model (LLM) brain. You interact with it like you might a friendly colleague, using plain, spoken language. With six cameras embedded in various locations and a Large Visual Model (LVM) component, it can see and understand the world around it. Built-in microphones and speakers give it the ability to listen, process information and speak in response. Fully electric, it can work for 20 hours without a charge.
Just a few weeks before Figure 02, Boston Dynamics released another short, wordless video showing its prototype Atlas. It’s waking up from what looks like a deep sleep, flipping its legs over and onto itself into a standing position in a Humanly impossible manner. It’s face, just a black, circular void.
And the race is on. Tesla’s Optimus was first released in August 2021, equipped with its own neural network and computer vision technology[i]. Canada’s Phoenix is another example, Phoenix humanoid robot is the sixth generation of solutions from Sanctuary AI [ii]. German robotics company Nurea has an entire range of humanoid and non-humanoid helpers designed specifically for the care industry[iii].
Meanwhile, China’s robotics market is experiencing an explosive renaissance. Healthy Loong is one example; 185 cm tall, weighing 82 kg, enjoying 43 degrees of freedom, thinking with 400 TOPS (Trillion Operations per Second)[iv]. And that’s only a small fraction of the number of humanoid and non-humanoid robots coming online globally right now.
Most robots shuffle along at around 1 meter per second and can carry about 40 kg. They can slowly and painfully arrange objects on a table, boil and egg or transport boxes. They have difficulty with latency (time needed to process language commands), navigating stairs or dealing with small obstacles (including being pushed over by Humans)[v]. But compared with one year ago, these advances are astonishing. We can already see the potential, and now it’s just a matter of time and engineering to get there.
Why now?
After a near 50-year robot winter, why are the robots suddenly coming online now? Why, like unembodied AI, does each new model release feel lightyears ahead of its predecessor? The future we dreamed of as kids – Humans zipping around in flying cars, helped by friendly, patient robots, capable of supporting us with any manner of household tasks, equally able to launch themselves into burning buildings to rescue the family cat?
It turns out that this fantastic machine we call the Human body actually takes an unfathomable amount of computational power to carry out the smallest movements. The number of tiny movements required to scramble an egg is well beyond our ability to manually describe with code.
The answer is AI. The current explosion of this technology has finally given the field of robotics the essential reasoning and compute it had been waiting for. AI learning and a related phenomenon called “Recursive Self Improvement” (RSI) describes the way an early system can identify errors in its own operating code and improve it, becoming smarter, and better, without Humans programming, testing and manual coding.
AI needs high quality data to work, and until now there wasn’t much organized, structured data showing how Humans move in a typical environment. Meta recently released HOT3D, a dataset built specifically to support 3D hand-object interactions[vi]. It contains over 800 minutes of ‘egocentric video’ – Humans doing ordinary, everyday things using a set of recording VR glasses; picking up objects, making coffee, feeding the cat, ironing shirts and cutting vegetables – from multiple perspectives as well as high-quality 3D pose annotations of hands and objects.
Robots learn to interact with the world thanks to a technique called “Telepresence”. A Human operator will hook herself up to a recording VR headset and cumbersome tracking suit called a Motion Capture Rig and goes about performing basic tasks. The robot wired to the other end of this can see and feel what she does, collecting every micro movement she makes, to add to the learning dataset. The robots start by mirroring these actions one-to-one, alongside their Human operator. Eventually they act alone, self-correcting along the way, until it has a set of “perfect” basic core functions it can build from. It takes a robot about ten hours to learn how to make a cup of coffee, open a bottle of wine or whip up pancake batter[vii].
Another approach is Physics Learning through Autoencoding and Tracking Objects (PLATO)[viii]. This system learns physics by analyzing simulated videos of objects interacting according to physical laws, distinguishing between realistic and impossible scenarios. We have all seen these mesmerizing simulations of endless robots walking over pixelated pyramids, running into one another, picking themselves up and walking on.
What for?
In 2022, Amsterdam’s Schiphol airport lost nearly 190 million euros due to a crisis in staff shortages, having to close down runways and disappoint thousands of travelers[ix]. In the third quarter of 2023, three quarters of Dutch businesses experienced an acute personnel shortage[x]. In 2023, Japan’s labor shortage was felt in 51% of businesses, leading to the closure of more than 300 companies[xi]. In 2024, Singapore has 194 job vacancies for every 100 available candidates; while in the US, there are 1.5 jobs vacant for every unemployed person[xii].
We find ourselves in an increasingly complex world with a shortage of labor. There are many reasons for this; an aging population, increased demand across all sectors. A complex world demands highly skilled workers. Manual labor is the primary driver of goods and services prices, accounting for about 50% of global GDP.
It doesn’t require much creative thinking to see where all of this is heading. As robots join the workforce, from factories to fisheries to farmland, the cost of labor will decrease until it becomes equivalent to the price of renting a robot. With increasing demand, the cost of robots will decrease until it will become a business necessity to replace all Humans at the manual labor level with these shiny, new workers.
The Business Case for Embodied AI
In January 2024, Figure AI began a collaboration with BMW to deploy robots its automotive manufacturing facilities; it found that the Figure01 robot was able to autonomously complete many difficult, unsafe or tedious Human tasks[xiii].
Embodied AI will inevitably work its way into a variety of industries, from factory labor, currently carried out by more than 3 billion Humans, to assisting individuals in the home (about 2 billion), to caring for the elderly (1 billion), and, eventually, for interplanetary exploration.
Industries in the line of sight include manufacturing, shipping and logistics, warehousing, and retail, where labor shortages are the most severe. At first, humanoid robots will take on structured and repetitive tasks, but over time, be able to handle more variable roles.
Tomorrow
The near future will be a time of displacement and disruption. As the robots enter the workforce, the Humans will be priced out. There will be widespread layoffs at the lowest levels of the economy – the levels least able to weather the disruption – which can only be helped by coordinated, massive and concerted global reskilling of these Humans for their next roles. This requires investment and vision as well as cooperation; none of which we Humans do very well.
But with these growing pains behind us, a transition to an embodied AI workforce will effectively “level up” the Human workers to become managers, strategic thinkers and planners, rather than do-er’s. A far better use of Human creativity and ingenuity and a more fulfilling working life. Humans will no longer have to do unsafe and undesirable jobs, but rather those that require finely-honed Human skills of connection, communication, empathy.
Soon, we will be sharing our homes and offices with the first generations of Embodied AI. They will be working in factories, hospitals, elderly homes, schools and offices. We will see them in our private lives, on construction sites, in our police forces and at war.
Are we ready to embrace this new strain of non-Human intelligent worker? What sort of worker laws apply to them in their pre-AGI state, and what will their status be post -Singularity? Will they be second-class citizens, or worse, slaves, to Humans? Have we simply created a new worker who will eventually demand rights, freedom and inalienable protections, such as the Right to a Power Source, Right to Data, privacy and the pursuit of robot happiness?
What a fascinating time to be alive.
Need help with AI Integration?
Reach out to me for advice – I have a few nice tricks up my sleeve to help guide you on your way, as well as a few “insiders’ links” I can share to get you that free trial version you need to get started.
No eyeballs to read or watch? Just listen.
Working Humans is a bi-monthly podcast focusing on the AI and Human connection at work.
About Fiona Passantino
Fiona is an AI Integration Specialist, coming at it from the Human approach; via Culture, Engagement and Communications. She is a frequent speaker, workshop facilitator and trainer.
Fiona helps leaders and teams engage, inspire and connect; empowered through our new technologies, to bring our best selves to work. She is a speaker, facilitator, trainer, executive coach, podcaster blogger, YouTuber and the author of the Comic Books for Executives series. Her next book, “AI-Powered”, is due for release soon.
[i] TechNode Feed (2024) “Optimus debuts today at Shanghai’s 2024 World AI Conference” TechNode. Accessed August 8, 2024. https://technode.com/2024/07/04/teslas-second-gen-humanoid-robot-optimus-debuts-today-at-shanghais-2024-world-ai-conference/
[ii] Global update (2024) “Sanctuary AI Unveils General Purpose Humanoid Robot” Global Update YouTube. Accessed August 8, 2024. https://www.youtube.com/watch?v=Q10A8Zkfd_M
[iii] Neura Robotics website (2024) “NEURA to Advance Cognitive and Humanoid Robots Using NVIDIA Isaac” https://neura-robotics.com/nvidia-and-neura-robotics-partner-to-advance-cogntive-humanoid
[iv] Chang (2024) “Shanghai AI conference stuns with 45 cutting-edge robots, dragon-themed open-source humanoid” Digitimes. Accessed August 8, 2024. https://www.digitimes.com/news/a20240709PD212/ai-robot-waic-2024-china.html
[v] Wodecki (2024) “Life-Size Humanoid Robot With Dexterous Hands Unveiled at World AI Conference” AI Business. Accessed August 8, 2024. https://aibusiness.com/verticals/life-size-humanoid-robot-with-dexterous-hands-unveiled-at-world-ai-conference
[vi] The Decoder (2024) “Meta’s new HOT3D dataset could enable robots to learn manual skills from human experts” The Decoder. Accessed August 8, 2024. https://the-decoder.com/metas-new-hot3d-dataset-could-enable-robots-to-learn-manual-skills-from-human-experts/
[vii] Sahota (2024) “Embodied AI: The Next Frontier in Robotics and Human Collaboration” Neil Sahota. Accessed August 8, 2024. https://www.neilsahota.com/embodied-ai-the-next-frontier-in-robotics-and-human-collaboration/
[viii] Sahota (2024) “Embodied AI: The Next Frontier in Robotics and Human Collaboration” Neil Sahota. Accessed August 8, 2024. https://www.neilsahota.com/embodied-ai-the-next-frontier-in-robotics-and-human-collaboration/
[ix] Sajet (2023) “Schiphol maakte een zooitje van 2022: nu ook miljoenenverlies voor luchthaven” NH Nieuws https://www.nhnieuws.nl/nieuws/314930/schiphol-maakte-een-zooitje-van-2022-nu-ook-miljoenenverlies-voor-luchthaven
[x] CBS (2023) “Drie kwart van de ondernemers ervaart personeelstekort” Centraal Bureau voor de Statistiek https://www.cbs.nl/nl-nl/nieuws/2023/34/drie-kwart-van-de-ondernemers-ervaart-personeelstekort
[xi] Inoue (2024) “Japan’s labor crunch persists despite slight improvement” The Japan Times. https://www.japantimes.co.jp/business/2024/05/06/economy/labor-crunch-continues-in-japan/
[xii] Ebbers (2024) “why is there a global labor shortage?” Randstad https://www.randstad.com/workforce-insights/talent-acquisition/why-there-a-global-labor-shortage/
[xiii] Adcock (2022) “Roadmap To a Positive Future Powered By AI” Figure AI Master Plan, Figure AI Website, Accessed August 8, 2024. https://www.figure.ai/master-plan