"How exactly are robots going to enter our lives?"
This is arguably one of the most frequently asked questions in the embodied AI industry over the past few years. Some believe that robots should first learn to do housework and take care of the elderly, eventually becoming omnipotent home assistants. Others feel that instead of waiting for a humanoid robot that can do everything, it is better to let robots master one specific task first.
For instance, helping people take photos and shoot videos.
In July this year, Insta360 unveiled its long-term vision for a "camera robot" named Cameraman. The goal is to create a Camera Agent capable of autonomously perceiving its environment, finding shooting positions, framing shots, and tracking subjects, without being confined to any specific product form factor.
Simply put, what Insta360 aims to build is a camera that no longer requires constant human control and can complete shooting tasks on its own.
At almost the same time, Miaodong Technology, founded by former DJI technical veterans, launched Beni—a small robot with two wheels that looks somewhat like a pet. It can follow users, avoid obstacles, jump, and even shoot 4K videos.
Compared to Insta360's Cameraman, which is still in the exploration phase, Beni has already completed its crowdfunding. With early bird prices starting at $549 (approximately CNY 3,674), it is set to begin shipping in November. The domestic release plan is synchronized with the large-scale overseas delivery, meaning it is just:
one step away from truly reaching consumers' hands.
However, Insta360 and Miaodong are far from the only ones eyeing the "camera robot" market. DJI and HOVERAir have long been exploring autonomous tracking via drones. The Qiyuan T1 and Vbot's robot dogs are also taking on mobile photography tasks. Even HONOR has integrated a mechanical gimbal and an AI Agent into its Robot Phone. The difference lies in the approach: some are building robots specifically for photography, while others are trying to teach robots how to take photos.
As the saying goes, all roads lead to Rome. But in the historical process of embodied AI reaching everyday users, will "photography" be a narrow path or a broad avenue?
Making the Camera Move on Its Own
If a camera robot is understood as a device capable of autonomous movement and following users to shoot, then it has actually already entered the consumer market. It is just that few people used to call it a "robot" in the past.
The most typical example is the drone. Take the DJI Neo 2, for instance. This small drone, weighing only 151 grams, already supports palm takeoff and landing, gesture control, intelligent tracking, and omnidirectional obstacle avoidance. Users do not need to hold the remote controller at all times; they can have it follow them while running or cycling, or execute camera movements such as orbiting and pulling away.
HOVERAir, under Hover Technology, launched the X1 flying camera focused on automatic shooting as early as 2022, followed by the X1 Pro and X1 Pro Max. Compared to traditional drones, they emphasize their identity as "flying cameras" rather than "drones." With palm takeoff, automatic tracking, and preset camera movements, they remove flight operations from the shooting process as much as possible.
In a sense, these products already possess the basic characteristics of a camera robot: the ability to perceive targets, control their own movement, and autonomously complete shooting to a certain extent.
Of course, flight has its limitations. Battery life, noise, usage venues, and safety restrictions all determine that drones cannot cover all shooting scenarios. Especially in environments like indoors or crowded streets, robots that can move close to the ground may actually have the advantage.
This is also a crucial reason why Miaodong's Beni chose a dual-wheel-legged form factor.
Unlike drones that shoot from above, Beni can follow users on the ground, utilizing its dual wheels and joint structure to achieve turning, jumping, and posture adjustments. It can shoot 4K videos, such as pets running, children playing, or dynamic ground-level shots in skateboarding. It can also express a pet-like sense of interaction through various movements.
TechRadar was particularly impressed by Beni's interactive design, noting that its expressions, sounds, and rolling movements make it feel like a lively robot pet at the same time.
More manufacturers are beginning to explore this route. For example, the T1 from Qiyuan Robot not only adopts a transformable design that freely switches between dual wheels and quadrupeds, but can also connect to action cameras like Insta360, supporting voice-activated shooting, automatic tracking, and orbiting camera movements. In addition to autonomous tracking, obstacle avoidance, and carrying loads, the Vbot super robot dog can also mount an action camera to act as a mobile photographer during outdoor activities.
Compared to Beni, these two products are not specifically designed around photography. Photography is just one of the important tasks they can perform, and the robots themselves carry more functions. This difference is actually quite interesting.
Furthermore, HONOR's Robot Phone launched this year integrates a mechanical gimbal directly into the smartphone. Through AI subject recognition, automatic tracking, and pre-framing, the lens can proactively adjust its direction as the subject moves, and even execute some professional camera movements.
A similar example is the Obsidian Tail 2, which achieves subject tracking through an AI active gimbal design and also supports a highly referential gesture interaction design (for details, see the Lei Technology review "Obsidian Tail 2 Live Streaming Camera Review: It Is an AI Gimbal Camera, and Also a Camera Robot"):
The active gimbal of the Obsidian Tail 2 does not mean we can just place the camera in any random corner. For the convenience of single-person use, the Obsidian Tail 2 features gesture control capabilities: forming an "L" shape with the thumb and index finger controls zoom, and the "OK gesture" can start or stop recording. However, if you do not place the Obsidian Tail 2 facing you, the camera will likely fail to accurately recognize the "OK" gesture.
Cameras with tracking capabilities existed before, but no matter how smart the algorithms were, they could only adjust image parameters, not the physical position and angle of the lens. In a broad sense, whether it is the HONOR Robot Phone or the Obsidian Tail 2, both can be considered a type of "camera robot":
At the very least, they allow the camera to stop passively waiting for user input and possess a certain degree of autonomous physical execution capability.
As for Insta360, although the Cameraman does not yet have a confirmed independent product form factor, it has actually accumulated a wealth of related technologies, ranging from 360-degree cameras, the Flow smartphone gimbal, and the Luna gimbal camera, to the InstaFly 360 drone. Insta360's vision is not to build a robot of a specific form factor, but to enable the camera to understand user intent and autonomously complete shooting, editing, and even sharing.
It must be admitted that today's "camera robot" is hard to define by a single product form factor. It can fly in the sky, run on the ground, look like a robot dog, or even just be a mechanical gimbal on the back of a smartphone.
In a public interview, Wang Mengqiu, founder of Hover Technology, said, "A flying camera is just a means, not an end." Liu Jingkang of Insta360, when discussing the Cameraman, also mentioned, "What most people really want is not to operate a machine, but to get good photos and videos."
The camera robot is not a suddenly emerging new concept, but the result of the intersection of two technological routes: the continuous autonomization of imaging equipment and the gradual practical application of consumer-grade robots. Ultimately, from a narrow perspective of demand, the "robot" in "camera robot" can only be a means, not an end.
At the same time, however, the camera robot can represent a possibility for robots to enter human daily life.
Learning to Track and Shoot First, Then Omnipotence?
To many people, the ultimate form of embodied AI seems to be the humanoid robot.
Yang Shuo, CEO of Miaodong Technology, does not believe that all robots should develop towards a humanoid form either. Yang previously participated in and led the R&D of DJI's flight control systems, later earned a Ph.D. in Robotics from Carnegie Mellon University, and served as an architect for the Optimus humanoid robot project at Tesla.
He gave an example in a previous interview: If a user wants a robot to hold an Osmo and follow them around to shoot, why must we first build a bipedal humanoid robot that can hold the camera with its hands? A small cart, or even a simpler mobile device, could also accomplish the same task.
In fact, this batch of Shenzhen manufacturers, including Insta360, HOVERAir, DJI, and Miaodong, have a more practical focus. They all emphasize satisfying "user needs" and value commercial implementation.
Therefore, it is not hard to understand why Miaodong chose the current product form factor for Beni. It does not need to mimic the human body, nor does it need to master complex dexterous manipulation. As long as it can stably follow the user and adjust the shooting position, it has the opportunity to complete photography tasks.
There is another point: photography requires the robot to perceive its surroundings, identify shooting subjects, plan its own movement routes, and continuously adjust its actions based on actual conditions, but it does not necessarily require the robot to directly manipulate and alter external objects. This means that, compared to complex household chores like tidying up rooms, doing laundry, and cooking, photography can avoid some unsolved technical challenges such as dexterous hands and object manipulation to a certain extent, while still allowing the robot to truly enter complex and ever-changing real-world environments.
But this does not mean Yang Shuo is not optimistic about humanoid robots. On the contrary, Yang mentioned in his personal column that Miaodong Technology decided to pursue two product lines from its inception: "one is a more long-term humanoid robot, and the other is a smaller robot that is smaller in size, safer, and easier to enter real life first."
In fact, the advantage of humanoid robots lies in their versatility. The environments, tools, and various infrastructures of human life are mostly designed around the human body. If robots need to operate these tools and complete various different tasks, the humanoid structure naturally has its value.
But if only one specific task needs to be completed, the situation is different. For example, if a robot needs to follow a person to shoot, it does not need to have two legs and two hands, let alone learn to tidy up rooms, do laundry, and cook first.
The key also lies in being "smaller in size, safer, and easier to enter real life first." From the perspectives of technological maturity, price, practicality, and product interaction, a product form factor like Beni clearly has the opportunity to enter mass life earlier than humanoid robots.
Over the past decade, smartphone, camera, and drone manufacturers have been lowering the barrier to shooting. From autofocus and computational photography to intelligent framing and automatic tracking, and now to AI editing, more and more tasks that originally required users to do personally have been handed over to devices.
Developing along this route, the next natural step is to let the device decide how to shoot on its own, and even when to shoot.
In fact, in Liu Jingkang's vision, the core of the Cameraman is a Camera Agent capable of understanding user intent, autonomously selecting lenses, shooting positions, and framing, and ultimately delivering satisfactory images. In other words, the development of camera robots is not just about teaching devices to move, but also about gradually letting them understand the very act of photography itself.
The other route starts from the robot itself.
The problem robots face when entering daily life today, apart from cost and other factors, is more importantly "what they can do." Robots performing martial arts and dancing, making ice cream, coffee, and pancakes, or handling inventory and retail are still mostly aimed at businesses. Relatively speaking, "photography" is more personal, which has also been verified in consumer drones, handheld gimbal cameras, and action cameras.
Therefore, for robots, photography can be a core function or just one of many capabilities. Robots can be designed specifically for shooting, or they can simultaneously take on tasks such as following, accompanying, and carrying loads.
These two routes are not mutually exclusive. They have different starting points but are converging in the specific scenario of photography. This also reminds us that the implementation of embodied AI does not necessarily require waiting for an omnipotent general-purpose robot.
However, having features like automatic tracking and obstacle avoidance is not enough. For a camera robot to truly approach the level of a photographer, it also needs to understand shooting timing, scene changes, and even the user's preferences for the image.
Therefore, from the Qiyuan T1 to the Vbot super robot dog, they all choose to achieve tracking by mounting cameras via external interfaces, letting manufacturers like Insta360 and DJI support the actual "shooting" part. At the same time, the products combine following, accompanying, carrying loads, and other life services, thereby attracting consumer purchases.
These explorations have at least brought the consumerization of embodied AI one step closer. As for whether it will ultimately be a drone, a wheel-legged robot, or a further evolution of existing cameras and smartphones, that is not the most important issue. What truly matters is whether robots can complete tasks that previously required personal human involvement, or even tasks that could not be done alone, without adding too much usage burden.
This also leads to a more realistic question. The demand for photography certainly exists, but today's smartphones, 360-degree cameras, drones, and smart gimbals can already satisfy a considerable number of shooting scenarios. Do we really need a new robot?
Before the camera robot products are truly implemented, I do not have a clear answer yet.