Orchestrating mechanics, perception and control: Enabling embodied intelligence in humanoid robots

Abstract

Humanoid robotics has evolved from early bipedal locomotion studies to modern systems integrating neuromorphic intelligence. This review systematically examines nearly 300 research studies, identifying key advancements in biomechanical optimization, multimodal perception, motion intelligence, and intelligent interaction. Recent progress in biomechanical optimization through material-structure co-design has led to lighter, more adaptive robotic frameworks, improving energy efficiency, compliance, and mechanical robustness. Meanwhile, multimodal perception has significantly enhanced environmental understanding by integrating vision, force, and proprioceptive sensing, enabling robust scene interpretation and adaptive interaction. However, challenges remain in real-time sensor fusion and uncertainty handling, limiting performance in dynamic and unstructured environments. Advancements in motion intelligence are increasingly driven by frameworks that integrate model-based control with learning-driven adaptation, allowing humanoid robots to achieve greater efficiency, agility, and generalizability in motion planning and execution. At the same time, intelligent interaction has evolved with approaches such as imitation learning, shared control, brain-computer interfaces, teleoperation, and large models, strengthening the link between perception and action for seamless human-robot collaboration. While these innovations enhance adaptability and interaction efficiency, robustness in intent-driven decision-making and real-world deployment remains a key challenge. Commercialization efforts have accelerated the transition from laboratory prototypes to practical applications, particularly in industrial automation and assistive robotics. However, scalability, autonomy, and safety remain critical concerns, requiring further advancements in hardware efficiency, neuromorphic computing, and AI-driven architectures. By synthesizing theoretical insights with recent technological developments, this review provides a structured roadmap for advancing humanoid robotics toward real-world implementation.