Robots have become increasingly popular in recent years, but the prices of most robots have not yet dropped to a level where ordinary people can afford one each. Currently, even with players like Qiyuan entering the "personal robot" market, a fully functional and structurally complete humanoid robot costs at least tens of thousands of yuan, not to mention those priced at hundreds of thousands or even millions.
Recently, a research team at the University of California, Los Angeles (UCLA) presented a unique solution: a 1.8-kilogram bipedal robot with a bill of materials (BOM) cost of only about $400 (approximately CNY 2,677). It sounds almost magical—how did they achieve this? The Leitech Embodied AI Team (ID: leitech) conducted a study and found that the concept behind this product is even more interesting than the product itself.
![]()
A $400 Bipedal Robot, Driven Surprisingly by a Smartphone
Recently, a paper appeared on arXiv introducing a robot named PhoneBot. The authors of the paper are from the RoMeLa laboratory in the Department of Mechanical and Aerospace Engineering at UCLA.
In terms of specific parameters, the PhoneBot stands 483 mm tall, weighs 1.8 kg, and features 13 degrees of freedom (DoF), with each joint utilizing a servo motor priced at $23.9. Most importantly, it is powered by a smartphone. Specifically, the smartphone is mounted on the robot's torso, simultaneously serving as the robot's sensor suite, onboard computer, and interactive interface.
According to the introduction, the capabilities of this robot include walking, standing up, following humans, taking photos, and conversational interaction.
Driving a robot with a smartphone sounds incredible, but PhoneBot has indeed achieved it. As the most popular consumer electronic device today, smartphones feature extremely high integration, with built-in cameras, SoCs (System on Chip), screens, Wi-Fi, batteries, and more. All these components can be utilized by the robot and become part of its capabilities. Equipped with a smartphone, PhoneBot can use the camera for image perception, the chip for computation and inference, and the network to access data and cloud computing power.
Of course, the $400 BOM cost mentioned at the beginning does not include the smartphone. However, the phones used for testing PhoneBot are all older models, including the moto G 2025 and Samsung Galaxy A15, with second-hand prices generally not exceeding $100.
Therefore, even including the hardware cost of the smartphone, the total BOM cost of PhoneBot remains under $500, which translates to just over CNY 3,000, making it absolutely outstanding in terms of cost-effectiveness.
Additionally, attempts to drive robots with smartphones started a long time ago. As early as 2021, a project named OpenBot emerged, using an ordinary Android smartphone to support the following and navigation of a four-wheeled robot. PhoneBot extends this concept to bipedal robots. In fact, the difficulty is much greater than that of quadrupedal robots: bipedal robots rely on real-time balance for standing and performing actions. Achieving this is not as simple as just connecting a smartphone.
Letting the Smartphone Take Over the Robot's Perception, Computing Power, and Interaction
Simply put, the PhoneBot solution is: the smartphone is mounted on the torso, acting as the robot's eyes, ears, and brain. The team divided its tasks into three parts: one for locomotion, one for image recognition, and one for speech.
The locomotion part is the most critical. The Android app on the smartphone reads the gyroscope and accelerometer data from the phone while simultaneously reading the current angles of the thirteen motors on the legs. Combined with the commands issued by the user, it compiles a real-time status report of the device.
The smartphone also runs a pre-trained small model. Upon receiving the information, it immediately calculates the target position for each leg and sends the commands via Wi-Fi to a small conversion board, which then relays them to the motors. The robot's walking, standing up from the ground, and adjusting its orientation are all achieved through this mechanism.
The image recognition part is entirely completed within the smartphone. The rear camera of the phone is responsible for recognizing people and objects. Combined with orientation commands, the robot can follow a person or turn its gaze to a specified direction. Localization is also handled by it; the smartphone calculates its own position using camera and gyroscope data, eliminating the need for an additional localization device. The screen is used to display status and visuals, allowing users to control it directly on the phone.
The speech part is slightly more complex. The smartphone handles audio recording and playback, but the computationally intensive speech-to-text conversion and large model inference still rely on a laptop placed nearby. The results are then transmitted back to the smartphone for playback.
As for how it learns to walk, the PhoneBot team's approach is to train it on a computer first. They first calculate the actual torque of each motor on the robot, then let the robot repeatedly try and fail in a simulated environment, falling and restarting until it walks steadily. During training, the team also mirrored the left leg's movements to the right leg, training both sides simultaneously. These training sessions were repeated five times for comparison. With the mirroring technique, the robot learns faster and walks more stably.
As mentioned earlier, PhoneBot uses older smartphone models with average configurations, but its hardware utilization rate is not high when driving the robot. According to data provided by the team, the CPU utilization rate when the phone reads gyroscope data is between 8.5% and 13.4%; when calculating walking postures, it is between 6.3% and 10.3%; when recognizing people and objects, it is between 15.6% and 17.9%; and for the most hardware-intensive localization, the utilization rate is between 25% and 34.3%.
The project tests demonstrate that PhoneBot is not picky about smartphones. The robot simply utilizes the cameras, gyroscopes, computing power, and system ecosystem accumulated by smartphones over the years; the specific brand and model are not that important. The low cost of PhoneBot essentially boils down to squeezing every bit of hardware resources from the smartphone, using whatever is available and saving wherever possible.
Of course, there is another crucial point: this robot has no arms. Although called a robot, its core locomotion capabilities are limited to the two legs used for walking and performing actions. Moreover, it is very small in size and does not undertake complex tasks such as carrying loads or climbing stairs. Ultimately, it is merely a testing device in a laboratory.
![]()
Will Old Smartphones Become a Powerful Tool for Robot Popularization?
Smartphones are arguably the consumer electronic products with the shortest upgrade and iteration cycles, with hundreds of millions of phones discarded globally each year. These devices integrate cameras, screens, chips, and various sensors, and most are still fully functional, discarded simply because they are not new enough. Often, the final fate of these devices is to be dismantled for material extraction, which largely represents a waste of massive hardware resources.
The emergence of PhoneBot provides a new outlet for second-hand smartphones. The idle computing power and sensors can be utilized in such small robots. It also proposes a new concept: the robot's brain can be treated as a consumable, while the main body is the long-term component. With smartphones being replaced every two to three years, the discarded models can be perfectly inserted into the robot. In this way, robots can change their brains just like changing batteries, becoming modular devices.
Of course, the drawbacks of this solution are also very obvious. Smartphones are not inherently designed for robots, so thermal management, power supply, and other aspects need to be re-engineered; configuration differences among various smartphone models will directly affect the robot's operational performance. Moreover, the quality and stability of second-hand devices lack guarantees. Another crucial point is that smartphones ultimately cannot provide robust computing power. Although PhoneBot is driven by a smartphone, it still relies on an external PC to provide part of the computing power. Furthermore, PhoneBot itself is just a small, relatively rudimentary robot without arms.
Nevertheless, PhoneBot has proven that driving a robot with a smartphone is feasible, thereby providing a low-cost solution for the industry to draw upon. For the entire robotics industry, the significance of PhoneBot may not lie in its immediate market launch as a consumer hit, but rather in breaking the hardware entry barrier for bipedal robots. For a long time, the high costs of bionic robot R&D have kept countless developers, university student teams, and startups eager to enter the field at bay.
The architecture of PhoneBot, featuring a "cheap torso + second-hand smartphone brain," actually plays the role of the Raspberry Pi in the robotics field. It holds the potential to spawn a massive, democratized developer ecosystem, giving more people the opportunity to develop and train their own bipedal robots.
Furthermore, given the continuous upgrade in the computing power of edge AI chips, the current shortcoming of PhoneBot—its reliance on an external PC for complex speech processing—can be resolved in the foreseeable future. Today's latest flagship smartphone chips can already smoothly run edge large models with tens of billions of parameters, completing complex semantic understanding and natural conversations locally.
Imagine that in two or three years, when smartphones equipped with contemporary flagship SoCs flood the second-hand market on a large scale, the neural computing power and sensor precision they provide will far exceed those of the low-end older models used in current PhoneBot tests. At that time, a single smartphone will be sufficient to endow such small robots with stronger intelligent features.
From a business model perspective, a product form similar to that of the PC industry might emerge in the robotics field, namely the "barebone robot." Hardware manufacturers would only be responsible for selling standardized robot bodies, batteries, and power components, driving down the prices of underlying locomotion hardware; ordinary users would simply need to insert an old smartphone into the robot's body to act as the brain, and then download a dedicated app to activate the device.
This not only aligns with the global demand for environmentally friendly recycling of e-waste but also enables the true popularization of devices such as home companion robots and educational robots.