Stanford Robotics Authority Reveals Insights for the First Time; Experts Warn Against Repeating the Mistakes of a Technology Bubble

On the 5th, Asia University hosted the “International Symposium on Humanoid Robotics and Sovereign AI for Future Living,” bringing together leading scholars and industry representatives from Taiwan and abroad to explore the critical challenges of transitioning humanoid robots from laboratories into the real world. The symposium not only showcased practical applications ranging from deep-sea archaeology to hospital care, but also provided in-depth analysis of core technologies, including chip architectures, cognitive models, and cybersecurity trust frameworks.

One of the most eye-catching highlights was the live demonstration of real-world applications. The “iBao Nursing Robot,” jointly developed by China Medical University Hospital and Changlian Technology, has already been deployed within hospital facilities to provide wayfinding assistance, health education interaction, and support for standardized care services. Meanwhile, the quadruped robot ANYmal from Swiss company ANYbotics demonstrated remarkable real-time environmental sensing and autonomous response capabilities, underscoring its strong potential for inspection tasks and disaster-response scenarios.

In response to the recent global surge of interest in AI-driven robotics, participating experts expressed excitement over rapid technological advances, while also offering pragmatic caution. They warned the industry against excessive hype reminiscent of the dot-com bubble and emphasized the importance of complementary human–machine collaboration rather than the blind pursuit of full automation.

The symposium was presided over by Asia University President Dr. Tsai Chin-Fa, who noted that humanoid robotics has reached a pivotal turning point. Development has evolved beyond engineering design and algorithm optimization to encompass more complex dimensions such as data security, system resilience, mutual trust between humans and machines, and societal acceptance. Asia University is committed to leveraging international exchange platforms to connect global expertise and accumulate practical experience and capacity for the real-world deployment of intelligent technologies in Taiwan.

Chairman Wu Cheng-Wen of the National Science and Technology Council emphasized in his remarks that amid accelerating global aging and declining birthrates, robotics technology is no longer merely an industrial issue but a vital pillar for sustaining national economic vitality. He stated that Taiwan’s comprehensive strengths in semiconductors, ICT, and precision machinery position it ideally to develop a fully integrated intelligent robotics ecosystem.

Wu further outlined the government’s strategic direction, which will focus on service-oriented applications in hospitality, healthcare, logistics, and disaster response. The policy framework aims to promote a systematic approach—from key technology R&D and regulatory standard-setting to talent cultivation. Through public–private co-investment, the government will support startups and SMEs, while fostering a “sovereign AI ecosystem” built on local data and independently developed models. At the same time, Taiwan will invest in next-generation computing technologies such as silicon photonics and quantum computing to ensure it maintains technological autonomy and competitiveness in the global innovation landscape.

The symposium’s opening keynote was delivered by Oussama Khatib, Director of the Robotics Laboratory at Stanford University. In his talk, titled “Shaping the Future of Human-Robot Collaboration,” he presented the remarkable achievements of the humanoid underwater robot OceanOneK in deep-sea archaeology.

During both his lecture and the subsequent Q&A session, Professor Khatib revealed several key technical details for the first time. He explained that unlike outer space, underwater environments pose immense fluid resistance and pressure challenges. To withstand the extreme pressure at depths of 1,000 meters, his team innovatively adopted composite materials embedded with hollow glass microspheres to construct a lightweight yet pressure-resistant body. The robot’s arms were designed as oil-filled limbs, with an external compressor balancing internal and external pressure. This design enables both the delicacy required to handle artifacts and the structural resilience necessary for deep-sea conditions.

Addressing the issue of network latency in cross-continental remote operation, Khatib introduced an innovative “dual autonomy system” architecture. Traditional feedback loops can cause oscillations under latency. In contrast, the dual autonomy system establishes independent autonomous systems on both the operator and robot sides. Instead of exchanging real-time state data, the two systems share input commands, effectively overcoming signal delay and allowing operators located on different continents to control the robot with precision and stability.

Khatib also explored the concept of “digital immortality.” He argued that merely recording a person’s voice or movements is insufficient to reconstruct their personality. True digital immortality, he suggested, requires extracting from massive datasets the strategies and physical models by which the brain controls the body. Preserving this deeper layer of intelligence and skill represents the ultimate goal of safeguarding human knowledge.

Another keynote speaker, Hiroshi Ishiguro, Director of the Intelligent Robotics Laboratory at Osaka University and widely regarded as a pioneer of modern robotics, delivered a philosophically rich presentation titled “Avatar and the Future Society.”

Ishiguro emphasized that the future of society lies not only in building smarter machines but in leveraging avatar and teleoperation technologies to transcend physical, spatial, and temporal limitations. He envisioned an inclusive society in which individuals caring for children or elderly family members, or those with mobility constraints, could participate remotely in work, education, healthcare, and broader social activities according to their lifestyles. For Ishiguro, the ultimate value of technology lies in designing systems that enable broader participation—and avatar technology represents a crucial pathway toward that vision.

Chang Chih-Wei, Visiting Professor in the Department of Computer Science at Stanford University, presented a talk titled “From Walking to Thinking: Feedback, Memory, and Causal Reasoning in Embodied AGI.” He examined the deeper challenges facing the real-world deployment of robotics.

Chang noted that while humanoid robots have made substantial progress in locomotion and basic manipulation, the primary bottlenecks in real-world deployment are increasingly cognitive in nature. He identified three major challenges: contextual loss in long-horizon task planning, inaccuracies in internal world models, and biases arising when inferring causal relationships from observational data. He stressed that improving the safety and reliability of embodied AI cannot rely solely on pre-designed environments or human supervision. Instead, more robust feedback loops must be established so systems can detect and correct errors before they propagate—an essential step for embodied AGI to reach maturity.

In terms of hardware architecture, Wei Shih-Chun, Special Assistant Manager at Realtek Semiconductor Corp., presented a forward-looking perspective. He argued that for robots to evolve from possessing a “silicon-based mind” into embodied AI agents capable of surviving and interacting in the real world, core chip architectures must fundamentally shift.

Wei explained that current Vision-Language-Action (VLA) models often rely on expensive, high-latency, and energy-intensive “System 2” reasoning. Such an approach is not optimal for robots that require real-time responsiveness. He therefore proposed a hierarchical dual-system architecture combining “System 2 for high-level planning” with “System 1 for low-latency reflexive control.” In his view, System 1 should be hardware-centric, leveraging innovations such as Computing-in-Memory (CIM), charge-domain computing, and high-density analog 3D IC integration. By embedding decision-making units directly into joints, motors, and robotic skin, robots can achieve ultra-edge, low-power, low-latency “reflexive intelligence,” enabling intuitive, biologically inspired responses.

As robots become more integrated into daily life, cybersecurity risks also intensify. Cheng Pao-Cheng, retired Senior Research Scientist at the IBM Thomas J. Watson Research Center, issued a serious warning during the forum: without adequate protection, humanoid robots could become launchpads for hackers to infiltrate internal networks—or even pose physical threats to humans.

Cheng emphasized that “perfect security” does not exist. Instead, the industry should adopt a resilience-oriented mindset: assume that attacks will eventually occur, and ensure systems can continue operating, minimize damage, and recover quickly. He stressed that defense strategies must incorporate a hardware Root of Trust, Trusted Boot mechanisms, and Trusted Execution Environments (TEE) to isolate sensitive programs and data at the hardware level.

During the panel discussion, experts engaged in lively debate over industry trends. When asked whether cleanrooms should be fully automated with robots, Professor Khatib and other panelists agreed that although cleanrooms are physically controlled environments, maintenance and troubleshooting introduce unstructured complexity. Full automation in such contexts is extremely difficult and costly. A more pragmatic approach is to deploy robots onsite while enabling remote experts to perform delicate operations via haptic interfaces—combining human decision-making with robotic physical presence.

Professor Khatib also candidly noted that while AI has made remarkable advances in perception, physical robots involve far more complex challenges in control, robustness, and safety. The technology is not yet fully mature. He cautioned the industry against repeating the mistakes of the dot-com bubble by overpromising outcomes, urging instead a step-by-step approach to solving fundamental physical and technical constraints.

Live demonstrations at the symposium further illustrated how robotics is gradually transforming real-world workplaces. In response to projections that the world could face a shortage of 9 million nurses by 2030, China Medical University Hospital presented a practical solution in the form of the “iBao Robot.”

Project Manager Chou Ying of Changlian Technology explained that the iBao robot integrates voice interaction, image recognition, and autonomous navigation. It can assist with non-clinical tasks such as wayfinding, item delivery, and smart meal ordering. According to on-site data, the robot is expected to reduce nurses’ walking distances by 15,000 to 20,000 steps per day, enabling healthcare professionals to focus more on core patient care. To safeguard patient privacy, the system operates on local servers using dedicated medical models. Future plans include integrating Vision-Language Models (VLMs) to assist with nighttime ward rounds, proactively detecting patient falls and issuing alerts.

Also showcased was the quadruped robot ANYmal, developed by Swiss company ANYbotics and demonstrated by its Taiwan distributor. ANYmal exhibited autonomous mobility across complex terrain, highlighting its potential for inspection, infrastructure monitoring, public safety, and disaster-response scenarios—serving as a reliable surrogate for humans in hazardous environments.

Huang Kuang-Tsai, Director of Asia University’s AI and Quantum Research Center, concluded that for humanoid robots to become trusted partners in healthcare, manufacturing, and service sectors, advances in hardware must be accompanied by strengthened data governance, system security, and cross-disciplinary integration.

In his closing remarks, Asia University President Dr. Tsai Chin-Fa stated that with “Future Living” as its central theme, the symposium not only demonstrated a surge of applications—from hospitals to the deep sea—but also clarified technological boundaries and cybersecurity challenges through international dialogue. Asia University will continue to serve as a platform linking academic research, industrial implementation, and governance frameworks, with the aim of accelerating the safe, trustworthy, and socially aligned deployment of humanoid robots and sovereign AI in the real world for the benefit of society.

On the 5th, Asia University hosted the “International Symposium on Humanoid Robotics and Sovereign Artificial Intelligence for Future Living,” drawing leading academic and industry representatives from Taiwan and abroad.