The widespread adoption of embodied artificial intelligence, represented by humanoid robots, may not take very long, but several significant challenges must still be overcome. Min Weidong, a deputy to the National People's Congress and Dean of the Metaverse Research Institute at Nanchang University, expressed this view during a recent interview.
With years of experience in computer vision, virtual reality, artificial intelligence, and big data research, Min believes the AI industry is currently undergoing a qualitative transformation. He pointed out that AI is evolving from a virtual "brain" to acquiring a physical "body," as demonstrated by robots featured in this year's Spring Festival Gala. Embodied AI involves giving artificial intelligence a concrete carrier, which could take the form of humanoid robots or digital humans.
According to Min, for embodied AI to become commonplace, it must overcome two major hurdles: technological breakthroughs and application-driven scenarios. Most companies currently focus on developing specialized robots for vertical industries, and there is still some distance to go before mass-producing versatile, adaptive humanoid robots. Min emphasized that achieving this transition from "specialized" to "general-purpose" robots will require breakthroughs across the entire supply chain, including chips, electronic components, and core software.
Min noted that the challenges facing embodied AI development are comprehensive. On the software side, more powerful algorithms and large-scale models are needed for support. On the hardware side, humanoid robots place extremely high demands on materials, chips, and high-end sensors.
Min predicts that within the next few years, embodied AI humanoid robots are likely to achieve breakthroughs in two scenarios: companion care and industrial manufacturing. Companion robots emphasize simple actions and emotional interaction, while industrial applications require precise operations. These two areas have the most urgent demand and are also the easiest in which to validate models. He also cautioned that safety capabilities must be embedded throughout the entire development process of AI.