Figure 03 - a review

There is something uniquely strange about watching a humanoid robot walk into a room. Not roll in on wheels like a delivery bot. Not sit behind glass like an industrial arm in a factory. Walk. Balance. Adjust its posture. Turn its head as if it is actually aware of the space around it.

That first moment is the real experience of the Figure 03, built by Figure AI. Before you think about specifications, artificial intelligence models, or robotics breakthroughs, your brain is still trying to process something more basic. A machine that moves in ways that feel very human.

The Figure 03 can help in day-to-day tasks, for instance in cleaning dishes.

Humanoid robots have existed for decades, of course. Companies have been experimenting with them for years, from research labs to viral demo videos. But the difference with Figure’s approach is not just the shape of the machine. It is the ambition. Figure is not building a robot for demonstration. The company is trying to build one that actually works in the real world.

The Figure 03 is the latest step in that direction, and if the previous generation felt like a promising prototype, this one feels much closer to a product. Not a product you can buy for your home tomorrow, but something that is starting to resemble a piece of technology with a real purpose.

At first glance, the robot looks exactly how people imagine a humanoid robot should look. Two arms, two legs, a torso, a head like sensor module. It stands roughly the height of an adult human, which immediately makes it feel less like a machine and more like a presence in the room. The proportions are intentional. Warehouses, factories, and human designed environments are built around the human body, and the robot is designed to navigate those same spaces without needing everything to be redesigned around it.

Up close, though, the illusion of humanity fades quickly. This is not an attempt to build a lifelike android. There is no artificial skin, no expressive face. The robot is unapologetically mechanical. Panels, sensors, cameras, and joints are visible. The design feels more like industrial equipment than science fiction.

That turns out to be a good thing. The robot does not try to pretend it is human. It simply uses the human form as a practical solution.

Watching it move is where things become fascinating. Earlier humanoid robots often walked with a kind of stiff caution, like someone carefully crossing an icy sidewalk. The Figure 03 feels noticeably more confident. Its gait is smoother. The balance adjustments are quicker and more subtle. When it picks up objects or repositions itself, the movements feel less robotic in the traditional sense and more like deliberate physical actions.

It still is not perfectly fluid. There are moments where the mechanical nature of the robot becomes obvious. A slight pause before lifting something. A tiny recalculation in posture when shifting weight. But compared with the earlier generations of humanoid machines, the improvement is striking.

The Figure 03 can interact with humans, have conversations, etc. actions that where to this day impossible to effectuate by an AI robot (due to delay).

What makes the Figure 03 particularly interesting is the intelligence layer sitting behind those movements. The robot relies on a combination of perception systems, machine learning models, and environmental awareness to understand what it is interacting with. Cameras and sensors continuously map the world around it, while the software attempts to interpret objects, surfaces, and obstacles.

This is where things become both impressive and complicated. The robot can identify common items, pick them up, and move them with surprising competence. Boxes, tools, containers. Tasks that are simple for humans but historically very difficult for robots.

But the system is still learning. Real world environments are messy. Objects are not always placed neatly. Lighting conditions change. Surfaces shift. The robot can adapt to some of this unpredictability, but not all of it. Occasionally it hesitates, recalculates, or takes a slower approach than a human worker would.

In a controlled demonstration, the robot can appear almost magical. In a real warehouse environment, it becomes clearer that autonomy is still evolving.

The hands are worth mentioning because they represent one of the most difficult challenges in humanoid robotics. Human hands are extraordinarily complex tools. Replicating that dexterity mechanically is incredibly difficult. The Figure 03’s hands do not match human flexibility, but they are surprisingly capable. They can grip, reposition, and manipulate objects with a level of precision that would have seemed unrealistic for commercial robots just a few years ago.

Watching the robot sort items or move objects across a workspace can feel oddly satisfying. Not because it is faster than a human worker. Often it is not. But because you can see the layers of technology involved in every motion. Vision systems recognizing an object. Control systems adjusting grip pressure. Balance systems compensating for the weight shift.

And then there is the artificial intelligence side of the equation. Modern humanoid robots are not just mechanical systems anymore. They are increasingly tied to large scale AI models that help interpret language, understand instructions, and learn from data.

Figure has been working on integrating conversational and reasoning capabilities into its robots. The idea is that eventually a worker could simply tell the robot what needs to be done. Move these boxes. Sort these parts. Bring that tool. The robot interprets the command and translates it into physical action.

In practice, that vision is still developing. The robot can follow structured instructions, but the fluid back and forth communication people imagine with robots in science fiction is not fully here yet. Sometimes the robot needs commands to be simplified or repeated in slightly different ways.

Still, the direction is clear. Robotics and artificial intelligence are converging in ways that make machines feel less like programmable tools and more like adaptable systems.

There is also a strange psychological aspect to being around a humanoid robot. Even when you know intellectually that it is just a machine, the human shaped form triggers certain instincts. When the robot turns toward you, you notice. When it moves through a shared space, you instinctively step aside.

That reaction raises interesting questions about the future of robotics in everyday environments. If machines are going to work alongside humans, the design of those machines will shape how comfortable people feel around them.

The Figure 03 walks a careful line here. It is human shaped enough to navigate human spaces naturally, but mechanical enough that nobody mistakes it for a person.

Of course, none of this comes cheaply. Humanoid robots remain incredibly complex and expensive pieces of technology. The sensors, actuators, processors, and engineering required to make something like the Figure 03 function are immense. For now, these robots are aimed at industrial environments where they might help address labor shortages or handle repetitive physical tasks.

Whether they become widespread will depend on many factors. Cost, reliability, safety, and the speed at which the software improves.

Despite its latency problems due to the company's emerging AI capabilities, the Figure 03 model is able to mimick human movements precisely giving it abilities to do complex hand work.

The reality today is that the Figure 03 feels both impressive and unfinished at the same time. It represents a huge leap forward in humanoid robotics, but also a reminder of how difficult the problem still is.

The technology is not perfect. It is not magical. It is still learning.

But it is here.