From TIMEWELL
This is Hamamoto from TIMEWELL.
Looking for AI training and consulting?
Learn about WARP training programs and consulting services in our materials.
Robotics Technology Is Evolving Fast
Robotic technology has been advancing at a remarkable pace, and quadruped robots in particular are attracting growing interest for industrial, entertainment, and even home use, thanks to their locomotion performance and versatility. Among them, the Go2 Pro — developed by Chinese robotics company Unitree Robotics — is expected to hold broad potential, from personal use to industrial applications, thanks to its refined design and advanced AI capabilities.
This article draws on a 24-hour comprehensive test of the Unitree Go2 Pro to report in detail on the robot's capabilities and potential for business use — from verification of core functions like AI obstacle avoidance, all-terrain mobility, and autonomous tracking, through to a social experiment conducted at an actual dog park. What changes might the Go2 Pro bring to our businesses and our lives?
- AI Obstacle Avoidance and Autonomous Navigation: Go2 Pro's Indoor Adaptability
- Versatile Controls and Intelligent Tracking: The Go2 Pro User Experience
- Can It Coexist with Society? Go2 Pro's Real Reactions at a Dog Park and Future Prospects
- Summary
AI Obstacle Avoidance and Autonomous Navigation: Go2 Pro's Indoor Adaptability
For a robot to perform practical tasks, the ability to perceive its surroundings, navigate autonomously, and avoid obstacles is essential. To test the Unitree Go2 Pro's AI obstacle avoidance and autonomous navigation capabilities, tests were conducted on a custom indoor course to verify whether Go2 Pro could assess situations in real time based on sensor data, select optimal paths, and execute them.
Specifically, a custom obstacle course was set up to test AI obstacle avoidance, and outdoor stairs along with a gravel slope were selected as terrains that would be difficult to predict — to evaluate all-terrain mobility. These tests revealed the sophistication of the Go2 Pro's embedded AI, its sensor fusion technology, and the refinement of its real-time motion control algorithms.
First, the "Free Avoid Mode" was tested on the obstacle course. In this mode, the operator only provides a forward directive, and the robot itself uses AI to detect and avoid obstacles while navigating to the destination. Rather than relying on precise human piloting skill, the idea is to test the robot's own decision-making capabilities. The course included obstacles of various shapes and heights, plus a surprise "dinosaur" object, creating a multi-dimensional load on the robot's recognition and decision-making capabilities.
On the first attempt, Go2 Pro recognized the obstacles in the course and began navigating around them skillfully. But here the robot behaved in a way that "betrayed" the test team's expectations: rather than following the constrained path of the obstacle course, it chose to navigate through the open space beside the course — selecting the easiest available route. This is evidence that the AI is not merely avoiding obstacles, but actively prioritizing the most efficient and safe path. In a sense, it demonstrated the AI's "intelligence" — though the test objective was specifically to complete the complex obstacle course.
So the "exit" spaces beside the course were blocked off with additional obstacles. After that, Go2 Pro occasionally stumbled or scrambled a bit, but ultimately cleared the obstacles skillfully and reached the goal. This suggests that an advanced AI navigation system is embedded that can accurately perform self-localization and obstacle mapping even in complex environments, then plan and execute appropriate actions based on that information. The test left a clear impression that Go2 Pro doesn't simply repeat programmed movements — it possesses a degree of autonomous judgment.
Next, real outdoor stairs and a steep gravel slope were selected for outdoor terrain testing. The focus here was on the robot's mobility — particularly its balance maintenance ability on uneven terrain, and how AI assists its movement.
"Freewalk Mode" was used for these tests. In this mode, AI assists low-level motion control such as balance maintenance and leg placement, while the operator controls speed and direction with a joystick.
On the stair challenge, the reviewer attempted a bold approach — entering at full speed. The robot stumbled momentarily but did not fall, successfully completing the ascent. This demonstrates its ability to sense the load on each leg and the inclination angle in real time, precisely controlling joint torque to absorb impact and maintain balance. When climbing the stairs slowly in Freewalk Mode, Go2 Pro captured each step surely and deliberately, with none of the stumbling seen at high speed — an extremely stable performance.
On the gravel slope, the main challenge was that the gravel could shift underfoot, potentially preventing sufficient grip. This is a frequently encountered situation in real outdoor environments.
As with the stairs, the first test was a full-speed ascent: kicking up gravel and raising a cloud of dust while struggling to climb the hill. The footing was constantly unstable, slipping occasionally — but remarkably, it reached the top without falling. Then, slowing down to Freewalk Mode, the behavior changed completely: proceeding slowly but surely, as if checking each foothold for stability, without any of the instability seen at high speed. The AI appeared to be smoothly absorbing the effect of shifting gravel and maintaining balance.
These rough-terrain test results proved that Go2 Pro has high adaptive capability not just on level laboratory surfaces but across the diverse terrains of the real world. Notably, there appears to be a different control strategy for high-speed running versus low-speed walking, and the selection of the optimal operating mode for each situation seems to be what supports its terrain traversal performance. This versatility raises expectations for applications in security, inspection, and disaster response.
Versatile Controls and Intelligent Tracking: The Go2 Pro User Experience
In addition to its advanced locomotion performance, the Unitree Go2 Pro pursues user-friendly operability. The manufacturer provides multiple control interfaces for users of varying skill levels and diverse use cases: a smartphone app, a dedicated remote controller, and a distinctive "mini remote." The smartphone app enables intuitive touch controls and detailed setting changes; the dedicated remote is suited for users who prefer precise direct control. Of particular note among these control methods is the "AI Follow Mode" via the mini remote. This AI follow mode enables Go2 Pro to track the user autonomously — significantly improving convenience for hands-free operation and for moving alongside the user.
In testing, the reviewer followed the manual's instructions, attaching the mini remote to their right side and positioning themselves to the left of the robot. Double-clicking a specific button on the mini remote switched Go2 Pro into autonomous tracking mode, and the robot began following the reviewer's movements without direct piloting. This function would be very useful for carrying items while walking, or for accompanying the robot on patrol of a specific area. According to Unitree, AI Follow Mode has two tracking speeds — low speed (max 1.5 m/s) and high speed (max 3 m/s) — though these speeds were not rigorously verified in this test. However, when the reviewer moved, Go2 Pro was confirmed to track smoothly, maintaining a consistent distance without lagging or coming too close.
However, some limitations of AI Follow Mode also became clear. When the reviewer deviated from the specified positional relationship (user on the robot's left side) or made sudden direction changes, Go2 Pro occasionally showed confused behavior and sometimes automatically exited AI Follow Mode. This suggests that the current tracking algorithm may be optimized for specific relative positions and movement patterns. Tracking target recognition likely combines signals from the mini remote (possibly UWB positioning technology) with camera and sensor information from the robot body, but tracking stability may degrade with unexpected movements or in complex environments with many obstructions. Future updates implementing more robust person recognition technology and flexible tracking algorithms that handle diverse movements are anticipated.
An important distinction in the product lineup: the more affordable "Go2 Air" model does not support AI Follow Mode. This suggests that the sensors and processing power required for AI Follow Mode are either omitted or downgraded in the Go2 Air — something to carefully consider when purchasing, depending on whether this feature is needed.
Can It Coexist with Society? Go2 Pro's Real Reactions at a Dog Park and Future Prospects
Following the technical performance evaluation, a more practical and fascinating test was conducted: a social compatibility experiment at an actual dog park with real dogs. This attempt is highly instructive for exploring how cutting-edge robot technology interacts with existing ecosystems and social environments, and what challenges and possibilities emerge.
When introduced to the dog park, Go2 Pro immediately became the object of curiosity for the real dogs. Some dogs seemed slightly nervous about the unfamiliar mechanical presence, showing wariness — especially when the robot turned to face them. Some dogs kept their distance and watched cautiously, while others approached without hesitation, sniffing and circling around the robot.
In terms of locomotion, Go2 Pro was clearly inferior to the real dogs in agility. To compensate for its relative slowness, though, Go2 Pro can perform "tricks" — spinning in place, bowing, dancing-like movements. Interestingly, these tricks seemed to impress the owners of the dogs more than the dogs themselves. "Oh my gosh!" exclaimed one person.
Over time, most dogs lost interest in the robot and moved on to other activities. However, one dog kept barking at it throughout the entire time at the park. Another dog persistently nuzzled its nose close to the back of the robot, apparently sniffing it. What that dog learned from the robot's "scent" is unclear — but the sight of an animal trying to figure out this mechanical entity was fascinating.
This dog park test highlighted the challenges and possibilities that robots like the Go2 Pro face in coexisting with existing social environments, particularly alongside animals. In its current form, not all dogs appear willing to accept the robot, given its relatively slow movement and mechanical appearance. However, the way its trick functions entertained people suggests potential as an entertainment device, or perhaps a role as a human companion. Combined with AI follow functionality, interactive applications such as a cargo-carrying assistant or a guide to specific locations are conceivable. For public deployment, careful consideration of the people and animals nearby, along with rigorous attention to safety, will be essential.
The reviewer noted they had "a great time," while also referencing Unitree's manual statement that "this is not a toy" — emphasizing the importance of reading and understanding the manual before use. This represents the essential mindset for safe operation and getting the most from a robot with such advanced capabilities. The dog park experiment provided valuable insight into how robot dogs are received in real society and what kinds of interactions they produce. Alongside technical refinement, the path forward requires exploring how to apply these robots while also considering social acceptability and ethical dimensions.
The 24-hour comprehensive test of the Unitree Go2 Pro clearly illustrated the remarkable technical level that contemporary quadruped robots have reached, the concrete possibilities for practical application, and the challenges that remain. Through wide-ranging testing — AI obstacle avoidance, all-terrain mobility, intelligent autonomous tracking, and actual interaction with animals — it became clear that Go2 Pro is not merely a novel gadget, but an advanced robotics platform with expected utility across many domains.
The advanced navigation capability and balance control demonstrated on the obstacle course and outdoor rough terrain proved that Go2 Pro can operate with a degree of autonomy even in complex, unpredictable real-world environments. In particular, the path selection in Free Avoid Mode and the rough-terrain traversal in Freewalk Mode speak to the sophistication of the embedded sensor array and AI algorithms. These capabilities evoke applications in security and surveillance, inspection of hazardous areas, logistics support, and even disaster response.
The diverse control interfaces and AI Follow Mode have contributed greatly to improving user experience. The hands-free tracking function via the mini remote raises expectations for use as a personal assistant or collaborative robot for specific tasks. However, tracking stability has room for improvement, and future software updates are anticipated.
The dog park experiment raised broader questions about how robots will be received by existing society and ecosystems. The dogs' curiosity, wariness, and eventual indifference caused deep reflection on the nature of communication between humans and robots, and animals and robots. The fact that trick functions were well-received by humans but had limited effect on dogs suggests the importance of interaction design tailored to the target audience.
Overall, the Unitree Go2 Pro represents a milestone in quadruped robotics technology. Its versatility and extensibility give it the potential to create new value in business, research, education, and eventually in everyday personal life. But to fully realize that potential, continued technical refinement must be accompanied by efforts to improve social acceptance and establish ethical guidelines. This test represents an important step toward that future. The trajectory of Unitree and the quadruped robot market as a whole deserves close attention.
Reference: https://www.youtube.com/watch?v=UMrE2T_L0yo