全球自动化：长远视角：机器人——物理AI与工业机器人复兴的新阶段

The Long View: Robotics -- Physical AI and the next phase ofindustrial Robot Renaissance Robotics is a frontier of physical AI. For industrial robots, there has been a notableinflection in adoption — a renaissance — since around 2020 (Exhibit 4). Most recently,the progression of AI is ushering this Robot Renaissance into a new phase, elevating theindustry’s CAGR to low-teens and increasing the long-term TAM by many folds. Jay Huang, Ph.D.+852 2123 2631 Weibin Liang, Ph.D.+852 2123 2666 The original Robot Renaissance was characterized by a shift from pre-programmed, fixedpaths to real-time flexiblepath planning, hence enabling new applications includingmachine tending, palletizing, and smart welding. The next phase advances from pathplanning to complextask planning(Exhibit 1), unlocking applications that require “brain”functions, such as long-sequence, high dexterity tasks, handling soft materials, and deepmachine/machine or human/machine collaborations. Without these new applications,industrial robot industry growth would likely moderate to single digit. With both phasesof the renaissance, we forecast the ten-year CAGR to accelerate to 12% and sustain wellbeyond the next decade (Exhibit 2). One way to understand the long-term potential is tosee the huge variance of robot penetration across industries and processes (Exhibit 6,Exhibit 7), and that enhanced robot flexibility, through path and task planning, can narrowthe penetration gap. Dien Wang, Ph.D.+852 2123 2622 Physical AI is the enabling technology behind the new Robot Renaissance. It is not a newtype of “AI robots” but a multi-layer AI ecosystem around the robots (Exhibit 8 to Exhibit 10).It consists of 1) robots and their digital twins; 2) task/path planning software increasinglypowered by multimodal AI (the “robotic brain models”); 3) sensors that collect physical dataof the robot and its environment; and 4) a digital representation of the environment with theability to simulate interactions that obey real physics (the “world models”). This is parallel tothe humanoid robotics ecosystem we discussed recently (see here). The following points are critical to assessing the impact of physical AI on industry players,but they are often misunderstood. First, Physical AI expands robot functions but does notdisrupt robot makers. The additional “brain models” and “world models” do not replacethe high precision motion control algorithm embedded in the robots (Exhibit 8). Second,the “brain” and the “world” are two distinct layers served by different players (Exhibit 9,Exhibit 10). Third, demand for sensors, of both vision and non-vision such as tactile, willgreatly increase, to provide increasingly sophisticated input to both robotic task planningand the building of the “world”. Fourth, leading robot makers such as FANUC are extendinginto the “brain” layer, but also actively seek external collaboration there and in the “world”layer. It is in this light we understand its recent announcement of opening to ROS2 and thecollaboration with NVIDIA (link). In industrial robotics, the word “physical” in Physical AI is embodied in physical equipment(robots), physical environment, and physical data. Key beneficiaries of this trend are FANUC,Keyence, and Mech-Mind (not listed). DETAILS The original Robot Renaissance was characterized by a shift from pre-programmed, fixed paths to real-time flexible pathplanning, hence enabling new applications including machine tending, palletizing, and smart welding. The next phase advancesfrom path planning to complex task planning (Exhibit 1), unlocking applications that require “brain” functions, such as long-sequence, high dexterity tasks, handling soft materials, and deep machine/machine or human/machine collaborations.Without these new applications, industrial robot industry growth would likely moderate to single digit. With both phases of therenaissance, we forecast the ten-year CAGR to accelerate to 12% and sustain well beyond the next decade (Exhibit 2). One wayto understand the long-term potential is to see the huge variance of robot penetration across industries and processes (Exhibit6, Exhibit 7), and that enhanced robot flexibility, through path and task planning, can narrow the penetration gap. EXHIBIT 1:The original Robot Renaissance marked a shift from pre-programmed, fixed paths to real-time, flexiblepath planning, and the next phase progresses from path planning to complex task planning. EXHIBIT 5:3D-vision-guided robot is on a steadily risingadoption curve. EXHIBIT 7:The upside potential in the broad conventionalmanufacturing sectors is significant. Physical AI is the enabling technology behind the new Robot Renaissance. It is not a new type of “AI robots” but a multi-layer AIecosystem around the robots (Exhibit 8 to Exhibit 10). It consists of : 1.Robots and their digital twins.Industrial robots are the frontier of physical AI, as they are the core components thatinteract with the physical world. Both the