Recently, I was looking forward to starting a series to take a break from these hectic days we all too often experience. I stumbled upon "1923," a dramedy and western that portrays the hardships of that era, including Prohibition, droughts, and the early stages of the Great Depression.
There was a particular scene that caught my attention. It showcased a washing machine as a novel invention at the time. While you couldn't speak directly to it, you could give it specific washing instructions. It was clear that many were astonished by the potential impact of such machines, and there was a palpable fear of being replaced in household chores. Although the washing machine couldn't respond with a voice, the preset instructions were sufficient for it to carry out various washing-related tasks, acting somewhat like a home assistant.
Today, the advancement of AI has enabled various "machines" to begin to come to life. Commands are now issued through voice, and robots are able to perform not just specific tasks, but also to make broader decisions.
Recently, the tech startup Figure introduced a humanoid robot named Figure 01 to the market. This robot can carry on a conversation thanks to its integration with OpenAI, which has effectively utilized multimodal capabilities. You might wonder, however, what sets this robot apart from others previously released by Boston Dynamics or Tesla.
After witnessing what Figure 01 could do in a demo, I was truly impressed with the tasks it managed to execute, the conversation it initiated with a human, and its responsiveness to external stimuli. It’s also worth noting that during this demonstration, the robot operated independently without any external controls.
How has a startup, founded just in 2022, managed to gain such significant momentum? Who are the people behind Figure, and what are their actual objectives?
It was just a few weeks ago that NVIDIA announced Gr00t, which could signify that we are at the brink of a new milestone in AI-driven robotics.
But perhaps more importantly, how will this new landscape that is emerging impact us as humans?
Exploring Figure AI and Its Humanoid
Currently, robot-related sales are taking off, reaching unprecedented growth rates of up to 48% in 2022. This surge has occurred in a context of labor shortages, where the most feasible solution has been to turn to technology that can substitute for human workers. Consequently, it is now quite common to find robots in warehouse spaces or on assembly lines.
It is within this scenario that Figure AI was established by Brett Adcock. The company is dedicated to the development of humanoid robots. It has also managed to secure approximately $675 million in funding from major backers including OpenAI, Microsoft, NVIDIA, Jeff Bezos, Intel Capital, and other investors who see potential in this sector. Figure AI's primary objective, as Adcock has noted, is to integrate humanoids into the workforce in areas where there is a demand for labor in high-risk positions or jobs that are unappealing to humans. This initiative is underscored by the presence of over 10 million jobs in the U.S. that are considered unsafe or undesirable, compounded by the demographic challenges of an aging population which makes it hard to meet all employment demands.
How does Figure 01 operate?
What distinguishes this robot is that everything demonstrated was learned through a variety of tests and training it underwent beforehand. It's equipped with cameras and microphones, enabling it to recognize its environment and understand spoken language. This capability is thanks to the multimodal technology from OpenAI, which Figure 01 utilizes. This technology allows it to read texts and images captured by the camera, which are then processed to formulate a response, either through speech or action.
The schematic we have shows how the humanoid developed by Figure AI begins to process various information sources. Notably, with each interaction with its environment, it enhances its performance through learning.
Another important feature is the mechanical speed of Figure 01’s actions, which reach up to 200Hz. This means that after deciding what action to take, it sends instructions 200 times per second, equivalent to every 5 milliseconds. Although this reaction time is not yet on par with human response times (250 milliseconds), the movements of Figure 01 are noticeably more fluid and less mechanical.
In summary, Adcock’s team has successfully built a robot that exhibits significant features such as dexterity, mobility, and natural language abilities. While there is still room for improvement in response time, this aspect is expected to evolve in tandem with advancements in OpenAI’s model.
Boosting the Race for AI Robots
Currently, the robotics industry is experiencing an unprecedented boom, evidenced by the diverse array of robots that tech giants have launched into the market:
Digit, created by Agility Robotics and backed by Amazon with a $1 billion investment, is capable of performing various tasks similar to those of a human within warehouse environments.
Apptronik has designed a robot named Apollo, noted for its ability to handle merchandise within a warehouse. Additionally, this company is collaborating with NASA to develop another robot called Valkyrie, which is designed to operate in hazardous environments and potentially for space travel.
Tesla Optimus, designed by Elon Musk's team, is focused on performing various tasks. However, it was disappointing to observe in one demonstration that the robot was not as autonomous as expected but was being externally guided. This revelation led to skepticism among many experts, and Musk eventually admitted that Optimus does not yet have full autonomy, which was a disappointment.
Gr00t from NVIDIA is based on a fundamental general-purpose model for humanoid robots. It aims to help other robots understand natural language and emulate human movements by observing human actions. This will enable it to quickly learn coordination skills and other abilities, allowing it to adapt and interact with the real world.
Considering the various ongoing robot projects and technological advancements, the integration of AI is almost imperative. AI will be the key component that speeds up the learning process for robots, enabling them to move from controlled environments to real-world settings that involve more variables but can be smoothly managed.
Final Thoughts
Perhaps everything you've read so far might make you wonder whether robots, just as they have learned productive and positive tasks, could also use their capabilities to pick up bad habits from their training data.
On another note, with such substantial investment yet only a few videos to showcase, one might speculate whether these projects are feasible and will deliver the anticipated impact, or if they'll just end up as another startup that misappropriated millions from eager investors.
Finally, remember that unlike humans, robots don’t need breaks or sleep. They don't need vacations or to go on strike (at least for now). I believe that as soon as robots become more economically accessible, companies will be keen to adopt them.