At the 2026 World Robot Expo (WRC), held from August 19-23 at the Beijing Yichuang International Convention and Exhibition Center, a special session on "AI Large Models Empowering Robots and Embodied Intelligence Industry New Paradigms" took place alongside the main event. Li Dangdang, Partner and Vice President of R&D at StarSense AI, participated in a roundtable discussion focused on "Scenario Application Innovation and Industrial Chain Collaboration." Here is the transcript of his remarks.

The moderator opened by asking each panelist to introduce themselves and their companies in one to two minutes. Li Dangdang began his introduction, explaining that StarSense AI was established earlier this year and has been focusing on embodied intelligence operations since the beginning of 2024. The company specializes in first-person perspective data collection, differentiating itself through proprietary annotation models for data labeling. Their priority product offering is annotation services, which they aim to provide to upstream and downstream partners. Beyond services, they also independently collect and annotate data for delivery to model developers or hardware manufacturers, and they manufacture their own collection hardware. Li shared that he spent 15 years in the automotive industry, with the last eight years dedicated to autonomous driving, before joining StarSense AI in May. He co-founded the venture with CEO Song Bo, who previously worked at a robotics company, where Li handles R&D, Song handles products, and two scientist colleagues round out the founding team. He expressed pleasure in exchanging ideas with industry peers at the event.

The moderator then steered the discussion toward two core themes: first, how scenario-driven innovation can create robots that genuinely function in real industrial conditions, and second, how building a collaborative ecosystem can break down supply chain barriers to accelerate industry-wide scaling. Noting that both he and Li came from autonomous driving backgrounds, the moderator observed that autonomous driving and humanoid robots are two applications of physical intelligence with similar underlying technology stacks in perception, decision-making, and execution. He asked Li about the key differences between delivering in the robotics space compared to autonomous driving, given Li's experience leading complete technical architecture and mass production at Li Auto.

Li responded that he frequently discusses this topic with colleagues, including why he transitioned to this industry. In his view, autonomous driving represents the initial phase of embodied intelligence, with many similarities in the complete loop of perception, planning, decision-making, and execution. However, while a car is physically larger, it occupies a smaller technical circle than robotics. He explained that a car operates in essentially two dimensions—moving forward, backward, left, and right—whereas robots can perform actions across many more dimensions. From this perspective, robotics presents a more complex, more systematic engineering and commercial challenge than autonomous driving. Yet much remains similar: the three core elements of data, compute, and algorithms that defined the autonomous driving era still apply to embodied AI. He noted that while some are solving hardware and embodiment challenges, others are addressing compute and chip issues, pointing out that NVIDIA chips work well in cars but encounter problems in robots. Li emphasized that the advantage of embodied intelligence is that it starts from a much higher baseline than early autonomous driving, when onboard compute was less than 1 TFLOPs and rule-based approaches were the only option. Now the industry can begin directly with models rather than hand-coded rules. While the fields are comparable and share references, he stressed that embodied intelligence represents a fundamentally larger and different paradigm.

When the moderator pressed further on what distinguishes robotics from autonomous driving beyond technical similarities, Li cited the supply chain as a major difference. The automotive supply chain is mature and established, whereas robotics still involves many newly developed components. Some elements, like batteries, can be borrowed directly from the mature automotive sector. But dexterous hands, for instance, are still in a very nascent stage and require significant effort to develop. Li admitted that hardware challenges in this area are critical—robots with poor dexterous hands simply cannot perform well. He drew a parallel to the early autonomous driving days when companies went sensor-heavy, adding lidar, millimeter-wave radar, ultrasonic sensors, and UWB. The first phase is always about pushing the upper limits—adding more modalities, increasing model parameters, and scaling compute—until the ceiling is discovered. Only after identifying those limits can the industry shift toward engineering deployment and mass production.

The moderator wrapped up the discussion by noting that this was his fifth year attending WRC, having returned to Beijing in 2022 before the term "embodied intelligence" even gained traction. He recalled that in 2024, industry progress seemed slow, but this year, visiting the exhibition halls, he was struck by how rapidly all robotics companies had advanced, far exceeding his expectations from just a few years ago. He asked each panelist to predict what attendees might see differently at next year's WRC, particularly since this year's theme has centered on practical implementation and deployment.

Li offered his perspective, suggesting that robotics will first achieve meaningful results in a relatively niche and specialized domain. He noted that contemporary embodied robots must be distinguished from traditional industrial robots. His expectation is that within the next one to three years, demonstrations will give way to genuine applications. While most current showcases remain performative in nature, he expressed confidence that next year will see robots performing real, purposeful actions that deliver tangible value. When asked to predict which buzzwords might dominate next year's discourse—given that VLA has cooled slightly and World Models are currently the hottest topic, with implementation being this year's emphasis—Li responded optimistically. He predicted that next year the industry will be talking about profitability, with companies emerging that have established commercial loops and are generating actual profits.