1. Overview
Modern technology is advancing toward systems that make decisions and act on their own without human intervention.
Autonomous technology is now extending beyond software into the physical world, and robots such as quadrupeds and humanoids serve as a conduit for bringing that autonomy into real-world spaces. In fact, these robots have begun to move out of the lab and into industrial sites, and the related market is growing alongside them.
As the robotics industry expands, so do the security threats targeting these products.
Security incidents involving robots do not stop at data breaches; they can lead to accidents that directly endanger human life. This calls for particular attention and for more research into robot security.
To build expertise in robot cybersecurity and contribute to the growth of the robot security industry, 78ResearchLab has begun vulnerability research on the Unitree Go2 Pro quadruped robot, and this series documents that process.
In Series #1, we examine why this research is needed by looking at the growth of the robotics industry and vulnerability trends, and then take a look at the exterior and basic features of our research target, the Unitree Go2 Pro.
2. Robot Market Trends and the State of Robot Security
2.1. Growth of the Robotics Market
According to the International Federation of Robotics (IFR) global industrial robot statistics, the operational stock of industrial robots worldwide reached 4,664,000 units in 2024, up 9% from the previous year, and annual new installations totaled 542,000 units, exceeding 500,000 for the fourth consecutive year.
By region, Asia accounted for 74% of all installations. By country, China hit a record high with 54% of global installations (295,000 units), while Korea ranked fourth in the world with 30,600 units[1].
With the worldโs highest robot density (1,220 units per 10,000 employees), Korea is a world leader in the use of robots in manufacturing[2]. The fact that a country ranked fourth in annual installations ranks first in density gives a sense of just how automated Koreaโs manufacturing sector is.
However, compared with its robot density ranking, the growth of Koreaโs robotics industry itself has been relatively modest.
According to the Korea Institute for Robot Industry Advancementโs โ2024 Korean Robot Industry Survey,โ domestic robot industry revenue in 2024 was approximately KRW 6.1695 trillion, an increase of only 3.2% from the previous year, with manufacturing robots accounting for 50.4% of the total[3].
These figures suggest that while Korea actively โusesโ robots, the value created by its robotics industry has yet to enter a phase of explosive growth.
In response, the Korean government announced its commitment to fostering the robotics industry through the โ4th Basic Plan for Intelligent Robots (2024โ2028)โ in 2024, setting policy goals such as deploying one million advanced robots[4].
Korea is currently going through a transition aimed at raising the added value of its robotics industry, and in the process, the security of robots deployed in and imported into the country has recently emerged as a policy issue. As symbolized by its world-leading robot density and its goal of deploying one million advanced robots, Korea as the country that deploys robots most densely must also shoulder a wider attack surface, and this has become the basic premise of robot cybersecurity threat trends.
2.2. Robots Deployed in Diverse Forms
The growth of robotics is reflected not only in the number of units but also in the diversification of robot forms. Starting with robotic arms used mainly in manufacturing, the range of applications is steadily expanding to service robots that move through homes and cities, humanoids that mimic the human form, and quadruped robots that walk with ease over uneven terrain.
Naver's robotic arm "AMBIDEXโ: ์ค์์ผ๋ณด
โข
Industrial Robot: As defined by the IFR, an automatically controlled, reprogrammable, multipurpose manipulator that moves in three or more axes, used in manufacturing and production settings. Typical examples include robotic arms used for automotive welding, electronic component assembly, and logistics palletizing.
Hyundai Motor's mobility robot "MobEDโ: ๋์์ผ๋ณด
โข
Service Robot: A robot that assists humans or equipment outside of manufacturing. Service robots are divided into professional service robotsโsuch as autonomous mobile robots (AMRs) for logistics, medical and surgical robots, and hospitality and delivery robots and consumer service robots, such as household robot vacuums and lawn mowers.
โข
Humanoid: A robot designed to mimic the human body (two arms, two legs, and a torso) so that it can use the tools and environments built for humans as they are. Examples include Tesla Optimus, Figure 02/03, Boston Dynamics Atlas, and Unitree G1/H1.
Unitree's quadruped robot "Go2โ: ๋ก๋ด์ ๋ฌธ
โข
Quadruped Robot: A robot that walks on four legs. Because it can move stably on stairs and uneven terrain, it is used for industrial inspection, reconnaissance, and disaster response. The Unitree Go2, the subject of this research, falls into this category.
Investment by major Korean companies also supports this trend. In early 2025, Samsung Electronics made Rainbow Robotics, a humanoid and collaborative robot company, its subsidiary [5], and Hyundai Motor Group jumped into humanoid mass production through Boston Dynamics. In particular, Boston Dynamics retired its hydraulic Atlas and unveiled a new all-electric Atlas, announcing plans to deploy it at industrial sites [6]. These moves show that robots are spreading across many industries, going beyond specialized equipment for specific fields.
2.3. Robot Security Threat Cases
The spread of robots brings with it the spread of attack surfaces. Security threats targeting robots have been reported one after another in recent years. Starting with early experiments on surgical robots, the scope of targets has expanded to industrial robots and robot middleware, and more recently to commercial quadruped robots and household robots.
Year | Case | Key Points |
2015 | Surgical robot โRaven IIโ experiment (University of Washington) | Demonstrated that cyberattacks on a telesurgery robot could tamper with commands and disable the robot |
2017 | ABB industrial robot hacking demo (Trend Micro) | Made a robot drawing a straight line deviate by 2 mm, demonstrating the danger of subtle defects |
2019 | Survey of ROS exposure on the internet | Scanned IP addresses worldwide and found many robot hosts exposed without protection |
2024 | Malware attacks targeting the robotics industry (Recorded Future) | Detected numerous RATs aimed at stealing intellectual property in suspected state-sponsored attacks |
2025 | Backdoor analysis of Chinese-made robots (Axios) | Access channels built into robots โ raised supply chain trust concerns |
2025 | Unitree โUniPwnโ (Go2, G1, H1, etc.) | Took over robots with root privileges via Bluetooth; raised the possibility of worm-like propagation |
2025 | KISA security inspection of robot vacuums | Confirmed vulnerabilities in some Chinese-made products that exposed in-home video externally |
[Table 1] Representative robot security threats and vulnerabilities by year
Listing representative robot security threat cases from the past decade shows how the scope of targets has expanded from limited domains, such as robots used as specialized equipment, to service robots and humanoid robots.
2.3.1. Cyberattack Demonstration on a Telesurgery Robot
In 2015, a research team at the University of Washington demonstrated cyberattacks on โRaven II,โ a telesurgery robot built for research.
With just a single malicious data packet, they were able to forcibly trigger the robotโs emergency stop and disable it, and they could also intercept and tamper with the commands sent by the surgeon.
The researchers warned that while operators could correct erroneous movements during simple tasks like moving blocks, such attacks could have far more serious consequences during precision surgery, where lives are at stake.
That said, it should also be noted that this robot is a research platform not approved by the FDA, and that clinical robots use secure networks [7]. This case shows that attacks on robots can go beyond data leaks and lead to physical harm.
2.3.2. UniPwn
Disclosed in September 2025, โUniPwnโ is a vulnerability in Unitree robots discovered by researchers Andreas Makris and Kevin Finisterre in the Go2/B2 (quadrupeds) and G1/H1 (humanoids).
Unitree robots use Bluetooth Low Energy (BLE) communication for initial Wi-Fi setup, and BLE packets are encrypted when sent and received. However, the BLE encryption key is hardcoded, and simply encrypting a fixed string with this hardcoded key and sending it is enough to pass authentication, allowing an attacker to impersonate a trusted device.
The attacker can then send the robot a Wi-Fi SSID and password with commands embedded in them, and those commands are executed with root privileges, without any validation, while the robot attempts to connect to Wi-Fi.
On top of that, the vulnerability drew attention because it raised the possibility of a worm-like botnet, in which an infected robot scans for other Unitree robots within BLE range and automatically infects them, spreading without any user interaction.
The researchers attempted responsible disclosure to Unitree starting in May 2025, but because the response was inadequate, they switched to public disclosure in September [8].
2.4. Trends in Robot Cybersecurity Research
If real-world threat cases show the actual damage, the research community has been refining frameworks to explain and respond to that damage.
2.4.1. Security Vulnerability Research on ROS
ROS exposure study mentioned above chose to scan IP ranges worldwide to identify ROS master nodes rather than analyze a specific device. Instead of looking at individual vulnerabilities, it measured โhow many robots are actually exposedโ at the population level.
The findings reveal a structural problem: robot middleware was not designed with authentication and encryption in mind. They point to vulnerabilities that allow anyone to steal sensor data or manipulate physical robots [9].
2.4.2. Cybersecurity Research on Quadruped Robots
An academic survey paper on the cybersecurity of teleoperated quadruped robots (Mohammad Sabouri, University of Genoa, submitted February 2026) starts from the concern that while quadruped robots are increasingly deployed in safety-critical missions such as industrial inspection, military reconnaissance, and disaster response, the security of the communication and control infrastructure linking operators and robots has not been sufficiently examined.
The paper argues that existing robot security frameworks are not sufficient for quadruped robots because of their dynamic stability constraints, vulnerable windows in each gait phase, high kinetic energy, and heavy cognitive load on operators. In particular, in its classification of defense technology maturity (TRL), it points out a clear gap between communication protection technologies already deployed in the field (TRL 7โ9) and cognitive- and operator-layer defenses that are still experimental (TRL 3โ5) [10]. These findings show that security research specific to quadruped robots is still in its early stages.
3. Why This Research Is Needed
The robotics industry trends discussed above can be summarized into three key characteristics.
1.
Robots are rapidly spreading into industrial sites and everyday life.
2.
Attacks on deployed robots have already been confirmed through demonstrations and real-world cases.
3.
However, security research that reflects the unique characteristics of robots is still in its early stages.
This research begins as an effort to narrow the gap between the growth of the industry and the current state of security research.
3.1. How Robot Security Differs from Conventional IT Security
Attacks on robots do not end with data; they manifest in the physical world through motors and joints. The telesurgery robot case shows that a single packet can trigger an emergency stop and tamper with commands. On top of this, there is a structural problem: the real-time requirements of control loops make it difficult to apply security measures.
Alias Robotics, a robot cybersecurity company, points out that because a robot is a โsystem of systems,โ strong authentication and encryption slow down control loops, and since even a delay of around 100 milliseconds can lead to collisions, falls, or danger to human life, manufacturers tend to sacrifice security for speed [11].
3.2. Why We Chose the Unitree Go2
There are two main reasons we chose the Go2 as our first robot research target.
First, the Go2 is a commercial quadruped robot with multiple external connection paths, including a smartphone app, official and unofficial SDKs, and the manufacturerโs cloud. The more diverse the connection paths, the more interfaces can be used as attack vectors.
Second, the Go2 was among the models affected by UniPwn, which we covered above as a real-world threat case, giving us the advantage of being able to compare our analysis against published prior research. Starting with the Unitree Go2, we plan to gradually expand our research to a wider range of robot products.
4. Target
4.1. Unitree Go2 Pro
4.1.1. Whatโs in the Box
The Go2 Pro comes with the robot body, a battery, and a remote controller that can control some of the robotโs movements.
Default Position
Front view of the Go2
Battery
RC
4.1.2. Product Specifications
Unitree Robotics, Go2 product page and specifications [URL]
The Go2 is a quadruped robot released by the Chinese company Unitree. It is a consumer and educational product equipped with a 4D LiDAR and 12 joint motors, and is controlled through a dedicated app. Prices start at USD 1,600 depending on the model, and the Pro, the model studied in this research, costs USD 2,800. Detailed specifications are shown in the table below.
Item | Specification |
Dimensions (standing) | 70cm ร 31cm ร 40cm |
Dimensions (lying down) | 76cm ร 31cm ร 20cm |
Weight (with battery) | Approx. 15kg |
Material | Aluminum alloy + high-strength engineering plastic |
Voltage | 28Vโ33.6V |
Max operating power | Approx. 3000W |
Payload | Approx. 8kg (max 10kg) |
Speed | 0โ3.5m/s |
Max climbing height | Approx. 16cm |
Max climbing angle | 40ยฐ |
Base computing power | 8-core high-performance CPU |
[Table 2] Detailed specifications of the Unitree Go2 Pro
4.1.3. Firmware Version
The firmware version currently installed on the robot can be checked in the robot control app.
โข
Hardware: v2.0
โข
Software: v1.1.11
4.2. Identifying Physical Hardware Information
This part covers only the hardware information that can be identified by visual inspection such as the mainboard, memory chips, and communication modules mounted inside the robotโs shell, as well as the physical interfaces exposed to the outside. Analysis of the actual system architecture and runtime environment will be described in detail in the next part.
4.2.1. Removing the Top Cover
Remove the robot top plate
Removing the robotโs top cover reveals the mainboard and various physical interfaces.
4.2.2. Mainboard
The chipsets identified on the front of the mainboard are listed below. Most of the chips on the back of the board had their markings erased, using methods such as laser masking, to intentionally prevent identification.
Front of Mainboard and Main Chipset
Back of Mainboard
Category | Component | Spec Summary |
SoC (main) | Rockchip RK3588S2 | ARMv8 big.LITTLE, Quad Cortex-A76@2.4GHz + Quad Cortex-A55@1.8GHz (8-core), 8nm, Mali-G610 MP4 GPU, NPU approx. 6 TOPS |
PMIC | Rockchip RK806-1 | Integrated power management IC, 10ch Buck + 6 LDO, I2C/SPI control, Master role |
eMMC | BIWIN BWCTAKC11X64G | 64GB, eMMC 5.1 (3D TLC + controller), HS400 |
DRAM | BIWIN BWMZAX32H2A | LPDDR4X 32Gb (4GB), up to 4266Mbps |
WLAN+BT | LB-LINK M8852BU1 (Realtek RTL8852BU) | 2T2R dual-band Wi-Fi 6 (802.11a/b/g/n/ac/ax) + BT 5.2 combo, USB 2.0 interface |
4.2.3. Physical Interfaces
In addition, various external interfaces such as USB-C, UART, and Ethernet were found on the mainboard. These interfaces can be used to analyze what kind of communication takes place between the outside world and the robotโs internals.
MCU adjacent port
Ethernet port
Etcโฆ interface ports
5. Identifying Go2 Pro App Features
With the Go2 Pro, users can install an app on their phone to steer the robotโs movements or have it carry out specific commands, much like using a game controller.
5.1. Robot Control App Features
5.1.1. Home
This is the first screen you see after installing the control app on your phone, completing the initial connection process, and logging in. The features on the right side are as follows.
Search Robot
Select wireless mode
Before Robot-App Connection
After Robot-App Connection
โข
Course: Guides โ videos
โข
Guide: Manual
โข
Me: Album, app information, etc.
โข
Function: Offers various modes such as BenBen Dog and Lively Pet
โข
Device: Settings
โข
GO!: Opens the screen for actually controlling the robot
5.1.2. Device
Users can also perform system configuration tasks through the control app, such as changing various settings and modes and running updates.
Callibration
OTA Update
5.1.3. Function
This menu provides various operating modes for interacting with the Go2, such as triggering actions through gestures and defining movements and actions through programming.
Lively Pet
Programming
5.1.4. GO!
This is the real-time control feature. It brings together Remote Control and the LiDAR map from Function, along with the camera, so that users can control and operate the robotโs movements through the app.
Real-time robot control
6. Conclusion
In Part 1 of this series, we discussed why this research is needed by examining the size of the robotics market and trends in security research, both in Korea and globally. We also identified the externally exposed interfaces of our research target, the Unitree Go2 Pro, based on its basic specifications and the hardware information identified from its exterior, and explored the robotโs basic functions and other information that can be gathered through the robot control application.
However, from the outside alone, we cannot yet tell which internal components receive the commands sent from the control app and turn them into actual movements, or what operating system and services are running behind the exposed interfaces.
In the next part (#2), we will cover our attempt to extract the robotโs firmware and our exploration of its internals, as the groundwork for the Go2 Pro research in earnest.
7. Reference
โข
[1] International Federation of Robotics, Global Robot Demand in Factories Doubles over 10 Years. https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years
โข
[2] International Federation of Robotics, Robot Density Surges in Europe, Asia, and Americas (World Robotics 2025, 2024 data), 2026.04.08. https://ifr.org/ifr-press-releases/news/robot-density-surges-in-europe-asia-and-americas
โข
[3] Korea Institute for Robot Industry Advancement, โ2024 Korean Robot Industry Surveyโ (revenue approx. KRW 6.1695 trillion, +3.2%), summarized by Robot Newspaper (irobotnews), 2024. (in Korean) https://www.irobotnews.com/news/articleView.html?idxno=44544
โข
[4] Ministry of Trade, Industry and Energy, โ4th Basic Plan for Intelligent Robots (2024โ2028),โ excerpt of press release via KDI Economic Information and Education Center, 2024. (in Korean) https://eiec.kdi.re.kr/policy/materialView.do?num=247119
โข
[5] TOPDaily, โSamsungโs Rainbow Robotics, Acquired at a KRW 2 Trillion Premiumโฆ โPhysical AIโ Expectations High, but Earnings Put to the Test.โ (in Korean) https://www.topdaily.kr/articles/108930
โข
[6] Boston Dynamics, Boston Dynamics Unveils New Atlas Robot to Revolutionize Industry. https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/
โข
[7] Help Net Security, Researchers mount cyber attacks against surgery robot, 2015. https://www.helpnetsecurity.com/2015/04/28/researchers-mount-cyber-attacks-against-surgery-robot/ / https://arxiv.org/pdf/1504.04339
โข
[8] IEEE Spectrum, Unitree Robot Exploit (UniPwn), 2025. https://spectrum.ieee.org/unitree-robot-exploit
โข
[9] N. DeMarinis et al., Scanning the Internet for ROS: A View of Security in Robotics Research, arXiv:1808.03322. https://arxiv.org/pdf/1808.03322
โข
[10] M. Sabouri, Cybersecurity of Teleoperated Quadruped Robots: A Survey, arXiv:2602.23404, 2026. https://arxiv.org/pdf/2602.23404
โข
[11] Boannews, โSuspected State-Sponsored Attacks Targeting the Robotics Industry and Risks Inherent in Robotsโ (citing Recorded Future/Insikt Group analysis). (in Korean) https://m.boannews.com/html/detail.html?idx=140928