When Elon Musk predicted at Tesla AI Day 2023 in March that one day there will be more humanoid robots than men walking on this planet, his notion was laughed off as a bad joke, a distant dream, or the sci-fi fantasy of a tech boss given to weed-fueled ravings.
But some 12,000 km away in the eastern Chinese city of Shanghai, a 30-year-old entrepreneur took Musk’s words seriously.
Peng Zhihui, better known by his avatar “Zhi Hui Jun” and a tech influencer with 2.5 million followers on video-sharing website Bilibili.com, is the chief technology officer of Agibot (智元机器人), a general robotics startup.
He echoed Musk’s words half a year ago and said a human-to-robot ratio of 1/1 is the “foundation” for what he and his co-workers are doing.
First-ever humanoid from a 6-month-old startup
Agibot, which was incorporated only in February this year, made headlines by launching its first-ever humanoid called “Raise A1” on August 18.
Although the robot walked onstage with a limp, it apparently had sufficient locomotion to move around bipedally and support its own weight, while maintaining its balance.
Besides, Raise A1 also possesses fine motor skills like pouring water into a glass and breaking an egg.
In effect, according to Peng, the startup had rolled out the prototype of Raise A1 as early as June, and has since worked on its motion control algorithms.
At the August 18 launch event, Peng introduced the series of technologies and innovations behind it, ranging from self-developed joint module PowerFlow to a robot hand SkillHand, from software framework AgiROS to a large language model (LLM) WorkGPT that serves as the robot’s “brain.”
Take PowerFlow for example. Employing a quasi-direct drive solution, the joint module features a peak torque of more than 350 NM and weighs in at 1.6k kg.
Bent on cost reductions
Industry insiders have pointed out that joint modules make up roughly 40% of a humanoid’s production cost.
To slash costs, Agibot started out by approaching contract manufacturers, only to find that none met its criteria. Instead, it was compelled to develop the component on its own.
The sheer efficiency of the young startup has taken the robotic industry, including some of its rivals, by surprise.
From tests of external parts, to initiation of the R&D process, then to coming up with a working prototype, it took Agibot only about a month to go through all these procedures.
Such efficiency, coupled with Peng’s personal credentials, helped it win the favor of major venture funds within a short period of time.
The company has secured four fundraisers from big-name backers such as Baidu.venture, Matrix Partners China, CDH Investments and Gaorong Capital, business data search services provider Tianyancha indicates.
Mass adoption starts with an affordable product
Despite all the adulation coming his way, Peng remains modest, saying Agibot is still way behind Western counterparts when it comes to the technology.
But he believes China does have a crucial advantage that is unparalleled across the world: a resilient manufacturing sector and a complete supply chain.
His next focus will be on achieving cost reduction for Expdition A1.
The price of Atlas, the iconic acrobatic humanoid from Boston Dynamics, easily crosses the US$1 million mark and scares buyers away. Peng says this is not a “technical route” he wants to imitate.
Instead, he sets his sights on building a robot that can sell preferably millions of units to bring the price down to an affordable range.
The US$20,000 threshold
The definition of “affordable” may vary but he more or less sees eye-to-eye with Musk on the desired cost of a humanoid.
Musk’s estimate of the Optimus’s cost is no more than US$20,000. Peng’s is a little higher, at less than 200,000 yuan (US$27,440).
“The reason I insist on controlling the budget within 200,000 yuan is that humanoids produced at higher than this cost stand no chance of finding the buyers,” he said.
Given his observations, humanoids in advanced Western countries tend to interact more with people, as labor costs are high, making robots a meaningful substitute for humans.
In some industries like senior care and household service, where there is a persistent shortage of care-givers and companions, demand for such contraptions is even more acute.
Application first in manufacturing
By contrast, demand for humanoids in China currently comes more from the manufacturing sector in search of ways to automate workflows and enhance efficiency.
Agibot has announced plans to accelerate commercialization next year, with a fleet of mass-produced Raise A1 to be adopted first in smart manufacturing.
The company is already in talks with top-notch manufacturers to discuss the purchase of its robots.
Traditionally, industrial and collaborative robots dominate the smart manufacturing scenarios at factories. Many people believe humanoids are not built for such an environment, since they are not agile and versatile enough.
Peng thinks differently. Humanoids can be capable workers in production facilities, as long as they acquire high degrees of freedom, become sufficiently dynamic and possess AI skills scalable to multiple use cases.
From corporate to individual use
After its products survive the test in manufacturing, Agibot looks to deploy them for non-corporate purposes, assisting workers, researchers and families in miscellaneous tasks.
“The approach will be a circuitous one, starting with business-oriented application and shifting to the average user,” Peng noted.
Going forward, Agibot will steadily open source its platform to developers.
Chinese media reported that the startup has established an R&D center on some 2,000 sqm of land in Lingang of eastern Shanghai.
Agibot is expected to conduct real-life scenario simulations and collect the data for analysis.
The passion for humanoids has been fueling Peng’s entrepreneurial endeavor all along.
Part of his motivation is rooted in the fact that robots in the form of humanoids are more suitable for a built environment than their non-human-like peers.
“We humans are used to living in an artificial environment, so humanoids are more suited to the life and production scenes we are familiar with,” Peng explained.
The advent of LLM
Back in college, he took part in various robotic projects and competitions, acquiring early experience in robotics.
His subsequent stints at Oppo and Huawei, where he earned the nickname “Huawei prodigy,” provided the launch pad for his free-wheeling tech aspirations.
The explosive growth of LLM early this year has fired him up.
In his words, the maturity of LLM and some general-purpose AI technologies signals the industry is now at an inflection point, “with something big on the cusp of happening.”
Additionally, the deep integration of AI and robotics serves as the prerequisite of what Nvidia boss Jensen Huang calls “embodied AI.”
The power to be unleashed by their marriage is his favorite subject, Peng claimed.
Asked about the reason he chose to quit big corporations to join a startup, he said their foray into robotics is driven largely by considerations of the impact on their main business. End-to-end application of robotics is rare.
On the contrary, he felt free to try his hand on something “uncertain” and unlock his potential at a startup — a luxury he couldn’t get whether at Huawei or Oppo.