In this conversation, we spoke with Jing Yu, co-founder of PANDAG, about why commercial robotics often breaks down outside controlled environments and how real-world perception, operations, and service shape long-term viability.
Q: In physical robotics, failure often comes not from an inability to move, but from not knowing what to do next. Was there a specific field scenario where PandaG realized that traditional RTK-based navigation fundamentally broke down in complex commercial environments, and that perception had become the true bottleneck rather than execution?
A: Most physical robots today operate in relatively standardized environments. However, commercial robotic mowing involves much more complex outdoor settings, where mobility itself is often more challenging.
This is because we are not only trying to solve intelligent operation on relatively standard lawns. Under ideal conditions — such as large open areas with no obstructions — RTK is indeed the most cost-effective and efficient navigation method. But when we move into a broader range of commercial scenarios, such as solar farms with extensive structural occlusion, we find that RTK alone is no longer sufficient. In these real-world environments, only a fused navigation system combining LiDAR, vision, and RTK can reliably address the practical challenges.
Q: Commercial operators often say that 90 percent automation is still operationally unacceptable. In your view, what is the most misunderstood “last 10 percent” problem in outdoor robotics, and why does solving that margin matter more economically than improving headline performance metrics?
A: From our perspective, the remaining challenges usually come down to two areas: hardware adaptability and the completeness of the commercial operating system.
First, the machine itself must meet basic operational requirements across different terrains and different types of grass. This is the foundation of any commercial mowing equipment, whether it is autonomous or not — it must cut the grass properly.
Second, the operating system must provide stronger support for commercial workflows. This includes whether the system can support automatic recharging at a level that enables efficient, unattended operation; whether it can support all-day, including night-time, commercial operation while ensuring safety and accuracy; and whether it can provide services such as fleet management.
Commercial mowing customers tend to focus more on efficiency, sustainability, and service reliability, rather than purely on performance metrics.
Q: PandaG emphasizes steep slopes, high grass, and hazardous terrain rather than manicured lawns. At what point did the team decide that eliminating high-risk human labor, rather than optimizing cosmetic precision, would be the defining line between consumer robotics and true industrial automation?
A: Although PandaG is clearly capable of handling heavier and more demanding environments, we can still support fine lawn maintenance through different blades and cutting decks, as these are also common commercial requirements.
Through real-world deployment, we found that the inherent hardware characteristics of the platform, combined with the adaptability of the navigation system, allow PandaG to take on certain high-risk tasks that are traditionally performed by human labor.
Q: When speaking with municipal buyers or large contractors, do you position the G1 primarily as a labor replacement or as a force multiplier that reshapes how human crews are deployed? How does that framing change the way CFOs think about ROI versus business continuity?
A: We prefer to position G1 as a “force multiplier.”
Its heavy-duty capability does not only improve day-to-day operational efficiency, but more importantly helps commercial customers expand into additional scenarios. It enables them to find solutions in incremental markets and extend their existing business models, rather than simply replacing labor on a one-to-one basis.
Q: The modular design of G1 suggests a philosophy that goes beyond seasonal mowing. Was modularity driven more by engineering ambition or by observing how idle capital erodes equipment ROI in winter months, and where did the biggest technical compromises appear in building a year-round platform?
A: Commercial customers face more complex operating scenarios. Mowing is our most common and universal entry point, but at a more fundamental level, customers need equipment that works across all seasons.
Modularity allows the platform to better match user needs beyond just winter use. It also enables us to deploy different modules for different commercial scenarios, rather than limiting the product to a single function.
Q: You’ve worked directly across markets like Japan, Australia, and the US, each with very different landscape standards and risk tolerance. Have you found that customers in different regions define “a job well done” in fundamentally different ways, and how does PandaG encode those differences into a standardized machine?
A: The core differences are usually not between countries, but between different commercial scenarios.
For example, the definition of “a job well done” is very different for manicured lawns compared to steep slopes. Through extensive testing across different scenarios, we not only segment the hardware, but also develop differentiated navigation and operation algorithms, allowing us to build and iterate navigation models tailored to specific vertical scenarios.
Q: Some competitors are moving toward Robotics-as-a-Service models that bypass traditional dealers. Why did PandaG choose to build local service and dealer networks instead, and do you believe large-scale physical AI adoption ultimately requires humans to remain in the operational loop?
A: Mowing is an industry that places strong emphasis on local service and responsiveness.
For commercial customers, having the right product and algorithms is only the first step. Ongoing support and on-site deployment are especially important. In addition, commercial customers rarely view delivery as a one-time event — their operating needs tend to evolve over time.
Therefore, even though building local service and dealer networks requires more time and upfront cost, we believe it is necessary in order to better meet customer needs.
Q: As fleets of G1 units map terrain through LiDAR and vision, they inevitably generate high-resolution spatial data. Do you see PandaG evolving primarily as a hardware company, or as a long-term spatial intelligence platform where data becomes as strategic as machines?
A: We believe PandaG will evolve into a productive intelligent spatial platform.
Each commercial environment has its own vertical data and scenario-specific navigation requirements. By continuously iterating on both the data and the navigation algorithms, each scenario can develop its own optimized solution.
Q: Looking back at PandaG’s early assumptions, what belief about technology, customers, or markets turned out to be wrong in practice? If you could send a message to yourself at the beginning of this journey, what is the most counter-intuitive lesson you would insist on learning sooner?
A: One important lesson we learned is the need to stay committed to iterating within real commercial scenarios, even when, at the beginning, those scenarios did not appear to be fully validated by the market.
