One unified system can do multiple tasks because the AI understands the whole body
Em uma terça-feira de maio, uma empresa californiana transmitiu ao vivo oito horas de robôs humanoides trabalhando de forma autônoma em um depósito logístico — sem intervenção humana, sem pausas, sem falhas visíveis. O que estava em exibição não era apenas uma façanha tecnológica, mas uma reivindicação sobre o futuro do trabalho manual em escala global: que as máquinas estão prontas para assumir uma das maiores categorias de emprego do planeta. A humanidade já viu automações transformarem setores inteiros, mas raramente com tanta clareza sobre quem, exatamente, será deslocado — e em que velocidade.
- A Figure AI transmitiu ao vivo robôs humanoides completando um turno de oito horas sem qualquer supervisão humana, elevando a demonstração de laboratório ao nível de argumento industrial.
- O sistema Helix-02 unifica visão, toque e movimento em uma única rede neural que opera até mil vezes por segundo — uma arquitetura que elimina a fragmentação que tornava robôs anteriores limitados e dependentes.
- Entre 15 e 20 milhões de trabalhadores de logística no mundo inteiro enfrentam a perspectiva concreta de substituição, enquanto a Amazon já retreina mais de 700 mil funcionários em antecipação à automação acelerada.
- A Figure passou de produzir um robô por dia para um por hora em menos de quatro meses, sinalizando que a transição do experimento para a escala industrial não é mais uma questão de 'se', mas de 'quando'.
- Concorrentes como Tesla, Agility Robotics e Sanctuary AI correm na mesma direção, e a transmissão ao vivo da Figure funcionou como um disparo de largada — ou um aviso.
Em uma terça-feira de maio, a Figure AI ligou as câmeras e deixou o mundo assistir. Seus robôs humanoides — os modelos Figure 03 — cumpriram um turno ininterrupto de oito horas em um depósito logístico, guiados pelo novo sistema de inteligência artificial Helix-02. Nenhum técnico interveio. Nenhuma pausa foi editada. Os robôs pegavam itens, moviam, empilhavam e seguiam em frente.
O que diferenciou essa demonstração das anteriores foi a escala da afirmação por trás dela. A Figure não estava apenas exibindo uma máquina capaz de andar ou segurar uma caixa — estava argumentando que esses robôs podem realizar o trabalho repetitivo e fisicamente exigente que hoje emprega entre 15 e 20 milhões de pessoas no mundo. Operadores de esteiras, separadores de pedidos, empacotadores, repositores: uma das maiores categorias de emprego do planeta, quase inteiramente manual.
O Helix-02 representa uma mudança de paradigma. Robôs industriais tradicionais são programados para tarefas específicas, cada um com sua própria lógica. A abordagem da Figure trata o robô como um organismo unificado: uma única rede neural controla o corpo inteiro em tempo real, integrando visão, sensores táteis e movimento. A arquitetura opera em três camadas — raciocínio, percepção e estabilidade — sendo a última ativa mil vezes por segundo, funcionando como um cerebelo digital.
A aceleração industrial é igualmente reveladora. Em menos de quatro meses, a empresa passou de um robô por dia para um por hora. Sua fábrica, chamada BotQ, já produziu mais de 350 unidades. Isso não é mais um experimento de laboratório.
O setor de logística tornou-se o principal campo de testes para esse tipo de automação. A Amazon, maior empregadora do setor, tem 1,58 milhão de trabalhadores globalmente e já retreina mais de 700 mil funcionários para funções técnicas. No Brasil, opera mais de 250 centros logísticos. Concorrentes como Mercado Livre, Shopee e Magalu expandiram suas redes rapidamente — e todos observam a mesma onda.
A demonstração da Figure ainda não prova viabilidade econômica em escala. Os custos são altos, a tecnologia é nova, e ambientes reais apresentam variáveis que uma demonstração controlada não captura. Mas ela mostra que a barreira técnica — aquela que tornava tudo isso impossível há cinco anos — está caindo rapidamente. E a transmissão ao vivo funcionou como um sinal: pelo menos uma empresa acredita que a corrida está quase no fim.
On a Tuesday in May, a California robotics company switched on cameras and let the world watch what might be the future of warehouse work. Figure AI streamed a live demonstration of its humanoid robots—the Figure 03 models—working an unbroken eight-hour shift in a logistics facility. No human stepped in to guide them. No technician paused the tape to fix a problem. The robots picked items, moved them, stacked them, and kept moving, all of it orchestrated by a new artificial intelligence system called Helix-02.
What made this different from previous robot demonstrations was the scope and the claim behind it. Figure wasn't just showing off a machine that could walk or grip a box. The company was arguing that these robots could now do the repetitive, physically taxing work that currently employs somewhere between 15 million and 20 million people worldwide. That number includes the people who run conveyor belts, pick orders from shelves, pack boxes, and store merchandise in warehouses. It's one of the largest employment categories on the planet, and it's almost entirely manual.
The Helix-02 system represents a shift in how robots are built and controlled. Traditional industrial robots are programmed for specific tasks—one machine does picking, another does packing, a third handles stacking. Each requires its own code, its own logic. Figure's approach treats the robot as a unified organism. A single neural network controls the robot's entire body in real time, integrating what the robot sees, what it feels through its sensors, and how it moves. The company describes this architecture in three layers: a reasoning layer that plans actions, a perception layer that converts sensory input into physical commands 200 times per second, and a stability layer operating at 1,000 times per second to maintain balance and coordination—essentially a digital version of a human cerebellum.
A year earlier, Figure had demonstrated this unified neural network approach with just the arms and hands. Now Helix-02 extends it to the whole body, enabling the robots to navigate real spaces, walk continuously, and manipulate objects without constant supervision. The company has also accelerated its manufacturing dramatically. In less than four months, Figure increased production from one robot per day to one robot per hour. The company's factory, called BotQ, has already produced more than 350 machines. This is no longer a lab experiment. It's a company moving toward industrial scale.
Warehouse and logistics work has become the primary testing ground for this kind of automation, and for obvious reasons. The work is repetitive, standardized, and physically demanding. Amazon, the world's largest employer in this sector, has 1.58 million workers globally, most of them in warehouse operations. The company already runs some of its largest facilities with 18 kilometers of automated conveyor belts and the capacity to process more than a million packages in a single day. Amazon is also retraining more than 700,000 workers for technical roles as it accelerates automation. In Brazil alone, Amazon operates more than 250 logistics centers with over 36,000 direct and indirect employees. Competitors like Mercado Livre, Shopee, and Magalu have expanded their logistics networks rapidly in recent years, and they're all watching the same automation wave.
The Figure demonstration doesn't yet prove that humanoid robots are economically viable at scale. The costs are still high, the technology is still new, and there are countless variables in real-world warehouse environments that a controlled demonstration doesn't capture. But the demonstration does show that the technical barrier—the thing that made this seem impossible five years ago—is falling fast. A handful of companies, including Tesla, Agility Robotics, Apptronik, and Sanctuary AI, are all racing to bring humanoid robots into industrial work. Figure's livestream was a signal that at least one of them believes the race is nearly over.
Notable Quotes
Figure claims its AI functions as a unified neural network capable of controlling the robot's entire body in real time, integrating vision, touch, and movement— Figure AI company statement
Amazon is retraining more than 700,000 workers for technical roles as it accelerates automation in its facilities— Amazon operations data
The Hearth Conversation Another angle on the story
What exactly was Figure trying to prove with this eight-hour demonstration?
That a humanoid robot can work unsupervised in a real warehouse environment for a full shift. Not just perform a single task, but navigate, adapt, and handle the variety of work that a human does in that setting.
But robots have been doing warehouse work for years. What's different about this?
Scale and autonomy. Previous systems were highly specialized—a robot that picks, a robot that packs. Figure's claiming one unified system can do multiple tasks because the AI understands the whole body as an integrated unit. And it did it without human intervention.
The company says it can produce one robot per hour now. Does that mean warehouses will start replacing workers immediately?
Not necessarily. Production capacity and economic viability are different things. A robot still costs more than a human worker in most markets. But the trend is clear—as production scales up, costs come down. Amazon's already retraining 700,000 workers, which suggests they're planning for exactly this scenario.
What happens to those 15 to 20 million warehouse workers?
That's the question no one has a good answer for yet. Some will be retrained for technical roles, like Amazon is doing. Some will move to other sectors. But the speed of this change is faster than most labor markets can absorb.
Is this actually a breakthrough, or is it marketing?
It's both. The technology is real—the unified neural network controlling the whole body is a genuine advance. But the livestream is also a signal to investors and customers that Figure believes it's ready to move from experiments to actual deployment. The company's betting its future on that claim.
What would stop this from happening?
Cost, reliability in messy real-world conditions, and regulatory or political pushback. But none of those seem like they'll slow this down much. The economic incentive is too strong.