Humanoid Robots and Humanoid AI: Review, Perspectives and Directions

Abstract

This comprehensive review examines approximately 60 years of humanoid robot evolution and the emerging field of humanoid AI. The paper presents a systematic terminology framework and paradigmatic landscape spanning human-looking to human-like and human-level humanoids. Rather than pursuing mere physical resemblance, the authors advocate for developing "humane humanoids" that embody both functional and nonfunctional specifications integrating generative AI, large language models, and human-level intelligence. The work addresses the "humanoid humanity dilemma"—the challenge of instilling genuine humaneness into physically human-resembling robots—and proposes transitioning from appearance-focused design toward systems exhibiting human-level cognition and ethical reasoning across real-time, interactive, multimodal applications.

Key Contributions

• Systematic categorization of humanoid evolution across six stages: structures, senses, behaviors, functions, humanity, and intelligence
• Three-paradigm framework distinguishing human-looking, human-like, and human-level humanoids
• Comprehensive taxonomy integrating robotics, AI, human science, and social science into a unified ecosystem
• Functional specifications framework detailing eleven essential capability dimensions for humanoid design
• Nonfunctional specifications model addressing human likeness satisfaction, capability maturity, performance evaluation, and impact estimation
• Analysis of ~30 humanoid robots with comparative assessment of their AI capabilities and architectural approaches
• Future research directions emphasizing omnimodal perception-to-action modeling and virtual-real humanoid integration

Methodology

The authors employ a comprehensive literature review and comparative analysis approach. They systematically examine existing humanoid robots categorized by developmental stages and paradigmatic approaches. The methodology incorporates multi-dimensional taxonomy development, examining humanoids through technical (mechanical, electrical, biological) and social design lenses. The framework synthesizes insights from robotics, AI, cognitive science, and human-computer interaction disciplines to establish functional and nonfunctional requirement specifications for evaluating humanoid systems.

Results

• Current market projections indicate humanoid robot market growth to USD $38–243 billion by 2035 with 13.8–50% annual growth rates
• Only limited humanoids leverage generative AI or large language models; none achieves true human-level intelligence
• Three capability phases identified: naive human-looking humanoids with minimal AI, ANI-driven standalone systems, and GenAI-enabled networked humanoids
• Existing humanoids demonstrate substantial gaps between physical anthropomorphism and meaningful humaneness, attributed to the "humanoid humanity dilemma"
• GenAI and LLM integration enables unprecedented real-time, interactive multimodal capabilities previously unattainable
• Vision-language-action (VLA) modeling represents emerging frontier for translating perception into meaningful behavioral outcomes

Limitations

• No existing humanoids truly embody human-level intelligence or authentic human consciousness
• The "uncanny valley effect" persists as a significant barrier to user acceptance of highly realistic humanoids
• Limited implementation of emotional, intentional, and consciousness-related features in current systems
• Insufficient integration of human science and social science principles into design and development practices
• Lack of standardized evaluation metrics for assessing humanoid humanity and nonfunctional specifications
• Ethical, legal, and social challenges remain largely unexplored in practical deployment contexts

Source: Humanoid Robots and Humanoid AI: Review, Perspectives and Directions by Longbing Cao, Macquarie University