Takeaways from DEEP Robotics visit: Power and firefighting revenue intact, industrial verticals still exploratory pilots, according to management

Research evidence excerpt

Takeaways from DEEP Robotics visit: Power and firefighting revenue intact, industrial verticals still exploratory pilots, according to management

acoustic

and temperature sensing; firefighting and patrol units are also being rolled out across

provinces including Hunan, Shanxi and Henan, alongside overseas park-and-warehouse

projects in Singapore and North America. The principal constraints, based on our external

industry survey, remain price and performance; while endurance, repeat-positioning

accuracy, multi-floor navigation and stability in dust, damp or extreme-temperature

conditions all remain only partially resolved. The survey further suggests that self-sensing

"smart-instrument" solutions could, over time, gradually erode the substation-inspection

opportunity as substations digitalise; the survey therefore sees power and firefighting as

credible near-term anchors, yet view broad, recurring industrial adoption as still

dependent on further cost-down and capability gains.

Physical AI still takes time as world-model paths remain unsettled

Management frames the sector's long-term trajectory around the "brain." It argues that

embodied robotics ultimately requires physical AI as a general-purpose technology, a

maturation it expects to take at least eight to ten years or longer, with today's models still

at an early stage. Crucially, the high-level technology roadmap has not yet converged:

leading academic work has been shifting from VLA-style (Vision-Language-Action) routes

toward world models, yet definitions of the "world" and scene understanding still differ

across players, leaving no firm timeline; by contrast, the "cerebellum," increasingly built

around RLMs (reinforcement learning models), is viewed as largely defined. On