The history of humanoid robots is an epic interweaving of human extraordinary imagination and relentless engineering efforts. It began with an ancient obsession with the ability to imitate human movements, undergoing centuries of conceptualization, prototypes, experiments, and setbacks. Only in modern times, with the explosive development of drive technology, computing power, and artificial intelligence, did humanoid robots gradually step from the laboratory into the real world. This chapter will trace this winding yet inspiring evolutionary path, analyze key milestones and their technological implications, and reveal the evolution of driving forces that shaped today’s landscape.
2.1 Early Concepts and Prototypes (Da Vinci, Early Automated Dolls) — Seeds of Dreams
The human desire for “artificial beings” has deep roots, embedded in mythologies (such as Hephaestus crafting golden maidens in ancient Greece, the Golem in Jewish folklore, and the wooden oxen in Chinese legends). The earliest attempts that truly embodied “humanoid” and “movable” prototypes were more concentrated in the field of mechanical automata, reflecting the outstanding understanding of kinematics by craftsmen of that time.
-
Leonardo da Vinci’s Concept (circa 1495):
-
In the Renaissance master Da Vinci’s manuscript, the Codex Atlanticus, there exists what is considered the first detailed design sketch of a mechanical knight.
-
This design included a clever transmission mechanism (driven by ropes and pulleys), intended to allow the “knight” to sit up, move its arms, turn its head, and even open and close its jaw. It resembled more of an exoskeleton platform demonstrating complex mechanical linkage rather than a truly autonomous, self-balancing robot, focusing more on showcasing the mechanical principles of human structure.
-
Although there is no evidence that Da Vinci actually built this knight, his drawings showcased early concepts of biomimicry and a structural understanding of humanoid movement, becoming a source of inspiration for future generations.
-
The Golden Age of Automata (17th-19th Century):
-
The Writer (1774): Created by Swiss watchmaker Pierre Jaquet-Droz and his son Henri-Louis and Jean-Frédéric Leschot. This complex mechanical device, resembling a three-year-old child, could actually dip a pen in ink and write custom phrases on paper (programmed through an internal complex cam mechanism), with its eyes following the writing action. Its arm’s writing motion was an early masterpiece of precise coordinated control of humanoid arms.
-
The Draughtsman (1774): Also from the Jaquet-Droz workshop, it could draw four different fine pencil sketches (of a ship, horse, dog, and couple), including actions of turning its head and “blowing away” pencil shavings. It demonstrated the ability to achieve complex trajectories through cam programming control.
-
The Musician (1774): A female child figure that could realistically play a small organ with sound pipes, with fingers pressing keys smoothly and accurately, simulating “breathing” with chest movements, and head and eyes moving naturally with the music. This was nearly the pinnacle of mechanical engineering at the time.
-
The Turk (1770): Created by Hungarian engineer Wolfgang von Kempelen. This famous device claimed to play chess automatically and defeat human opponents, but was later revealed to be a carefully designed hoax — a real human chess player was hidden inside. Although it was not truly automated, its realistic appearance (humanoid torso, arms, dressed in Turkish attire) and the sensation it caused profoundly influenced public imagination about the possibilities of “automated humans”.
-
Significance and Limitations of Early Explorations:
-
Driving Force: Primarily relied on clockwork, hydraulic pressure, gravity, or external steam engines for power, achieving preset, limited, single-step or sequential actions through precise mechanical structures like cams, gears, levers, and pulleys.
-
Intelligence and Control: Virtually nonexistent. Movements were based on fixed program open-loop control (like cams), unable to perceive the environment or make adaptive changes.
-
Form and Function: Mainly crafted as art pieces or entertainment devices, with single functions. Walking capabilities were generally absent or extremely rudimentary (like wheeled bases or shuffling legs).
-
Significance: These masterpieces are a testament to human engineering wisdom and artistry, achieving astonishingly precise mechanical movements (especially in arm and finger operations), providing inspiration for later engineering. They sowed the seeds of the dream of humanoid robots, but there remained a vast gap from truly autonomous, intelligent, and mobile humanoid robots.
2.2 Modern Pioneers: Milestone Platforms — Foundations of Reality
Entering the second half of the 20th century, with the rise of electronic technology, servo control theory, and computer science, true modern humanoid robots began to emerge. A series of milestone projects defined the direction and broke through key obstacles.
-
The WABOT Series (Waseda University, Japan, 1970s-1980s):
-
WABOT-1 (1973): The world’s first full-sized humanoid robot with a human-like appearance. It could walk slowly on a plane (requiring external guidance), use tactile hands to manipulate objects, and communicate remotely through limited visual and tactile sensors (recognizing about 50 Japanese words). Although bulky, slow, and reliant on external computation, its breakthrough lay in integrating perception, cognition (simple), planning, and execution within a humanoid framework, establishing the basic research paradigm for humanoid robots.
-
WABOT-2 (1984): Focused on musical interaction. It could “read” sheet music visually and play on an electronic keyboard with five mechanical fingers, with coordinated and smooth finger movements, significantly improving response speed. This demonstrated the potential of humanoid robots in specific task areas (fine manipulation) and was an important practice in motion coordination control.
-
The ASIMO Era of Honda (1986-2018):
-
E0-E6 (1986-1993): The experimental platform of the “E” series focused on basic walking technology research, evolving from slow static gaits (E0) to dynamic walking (E6), establishing Honda’s leadership in bipedal walking.
-
P1-P3 (1993-1997): The “P” series began to feature a complete humanoid structure (torso, head, arms, hands). P1 used hydraulic drive, P2 was the first humanoid capable of walking autonomously (without cables), climbing stairs, and balancing by grasping handrails, while P3 was miniaturized and improved in reliability. They were the direct technological foundation for ASIMO.
-
ASIMO (2000-2018): The most influential and publicly recognized early humanoid robot model (Advanced Step in Innovative Mobility). Each upgrade continuously improved its walking capabilities (from shuffling to running, jumping, and single-leg jumps), motion fluidity (like “pouring a drink”), human-robot interaction (gesture, voice recognition and synthesis, facial recognition), and even stair climbing and adaptability to different terrains (slopes, uneven surfaces). ASIMO represented an era — driven by high-precision servo motors and excellent real-time gait control algorithms, achieving highly dynamic stable motion. Although its commercialization was unsuccessful, its technological showcase greatly stimulated global humanoid robot research and validated many core control concepts.
-
The HRP Series of AIST (1990s-Present):
-
HRP-1 (1997): More inclined to serve as an open research platform rather than a perfect performance device (like ASIMO). It adopted a distributed control system architecture early on.
-
HRP-2 “Promet” (2002): A landmark platform that marked a turning point in openness. AIST chose to make its hardware and software details relatively public (like controller interfaces, simulation models). This gave rise to the OpenHRP simulator, significantly lowering the entry barrier for global research institutions. HRP-2 excelled in fall protection (structural design allowed safe falling and getting up) and environmental adaptability (capable of operating in real environments like offices, factories, and outdoors), promoting practical exploration of humanoid robots (like disaster response drills).
(Note: Some early humanoid robot research teams in China began algorithm development and tracking research based on HRP-2 in the mid-2000s.)
-
HRP-4 (2009): More lightweight, compact, and humanoid (especially the female design HRP-4C), emphasizing human-robot collaboration and operational capabilities. The subsequent HRP-5P (2018) focused on heavy object manipulation (like panel installation).
-
Atlas of Boston Dynamics (2013-Present):
-
Extreme Hydraulic Drive: A hydraulic system with ultra-high power density and response speed (highly integrated valves and actuators).
-
Nonlinear Model Predictive Control (NMPC): Millisecond-level dynamic control optimization under full-body dynamic model constraints.
-
Perception-Driven Real-Time Planning: Utilizing stereo vision and LIDAR to instantaneously generate foot placements and action sequences.
-
2013 Atlas: Built based on the requirements of the DARPA Robotics Challenge (DRC), primarily hydraulic-driven, with strong power and certain disturbance resistance, but walking was relatively slow and cumbersome, relying on tethered power and control.
-
2016 Atlas (DRC Version): Eliminated the tether, enhancing endurance and autonomy, demonstrating strong comprehensive environmental adaptability in the DRC, including driving vehicles, traversing rubble, clearing obstacles, opening doors, operating valves, climbing ladders, and responding to unexpected falls, although it still appeared somewhat clumsy during the process.
-
New Generation Atlas (2018-Present): Showcasing unprecedented dynamic movement capabilities of humanoid robots — backflips, high-difficulty gymnastic moves (“gymnastics choreography” video), precise jumps, and parkour-like navigation through obstacles. Its core lies in:
-
Significance: Atlas redefined the limits of dynamic performance and environmental adaptability of humanoid robots, showcasing the physical potential they could achieve, greatly stimulating the industry’s pursuit of extreme technologies. Although the complex movements demonstrated were pre-programmed or achieved in constrained environments, its control algorithms represent the highest level.
-
NAO & Pepper from SoftBank Robotics:
-
NAO (2006-Present): The most widely used humanoid robot for education and research globally (height ≈ 58cm). Relatively low cost, modular design, and an open software development environment (supporting ROS). Mainly applied in STEM education, human-robot interaction research, and rehabilitation training assistance. Its value lies in greatly expanding the developer ecosystem for humanoid robots, allowing many students and researchers to explore control, perception, artificial intelligence, and interaction algorithms at low cost.
-
Pepper (2014-Present): Positioned as an emotional recognition interactive robot (height ≈ 120cm). Targeting commercial scenarios (retail, reception). Its core lies in achieving multimodal interaction through a microphone array, cameras, and touch sensors (recognizing expressions, voice tones, and some emotional states), equipped with a tablet for information display. Although relatively weak in mobility and operational capabilities, its commercial attempts in specific scenarios (simple reception guidance, emotional engagement) provided important market experience (though not achieving expected success).
-
Tesla Optimus (Tesla Bot, 2022-Present):
-
Ultimate Engineering Design: Pursuing lightweight, highly integrated joints, using a large number of motors (including reducers) for actuation.
-
Reuse of Electric Vehicle Technology: Claimed to use self-developed motor/reducer units, battery technology (4680), sensors (primarily camera-based FSD vision stack, supplemented by IMU, etc.), and computing platforms (Tesla SoC).
-
End-to-End Artificial Intelligence: Relying on vision (primarily cameras), imitation learning, and reinforcement learning to train control strategies (actions, operations, navigation), aiming to handle unstructured tasks.
-
Launched by automotive giant Tesla, aiming to create a low-cost, high-volume, general-purpose humanoid robot for industrial and human use.
-
Its technological path emphasizes:
-
Significance and Controversy: Optimus represents a large-scale commercial entry by a new force (technology consumer electronics giant). Its progress is closely watched but has yet to be validated on a large scale. Its challenges include: Can it overcome the inherent complexity of humanoid robots to achieve sufficient reliability? Can it effectively transfer its claimed AI training methods to the physical world? Can it truly achieve the target cost of under $20,000? Its development will profoundly impact the industry’s resource investment and commercialization confidence.
2.3 Evolution of Driving Technologies (Hydraulic, Motors, Artificial Muscles) — The Source of Mobility
Driving technology is the core that enables robots to move and generate force. The complexity of humanoid robot movements requires actuators to possess high power-to-weight ratio (power/mass), high response speed, good control precision, and certain force density (force/volume or force/mass). Its development history reflects the path of technological breakthroughs:
-
The Rise and Refinement of Electric Motor Drives:
-
Series Elastic Actuators (SEA): Adding elastic elements (like springs) at the output of the reducer enhances impact force perception (used in PR series, early Atlas legs, and some research platforms).
-
Modular Integrated Actuators: Highly integrating motors, reducers, encoders, driver circuits, and even torque sensors into a compact unit (like Boston Dynamics’ new Electric Atlas joint unit, Optimus’ joint design is similar), reducing cables, enhancing power density and reliability.
-
Direct Drive Motor Exploration: Providing extremely high bandwidth and very low impedance for applications requiring extremely high dynamic response (like eye movements, high-speed balance adjustments), but with low torque density, not widely adopted in major load-bearing joints.
-
Hollow Cup Motors: Lightweight, efficient, and responsive, commonly used in small joints like fingers (like NAO, Pepper, and some dexterous hands).
-
DC Motors: Early foundation, facing issues like brush wear, low efficiency, and limited control precision.
-
Brushless DC Motors (BLDC): Became mainstream. High efficiency, high reliability, maintenance-free without brushes, and good control precision (with encoders).
-
Harmonic/Planetary Reducers: The high-speed low-torque characteristics of motors must be amplified through high-precision, low-backlash reducers to achieve precise positioning and significant output force, making harmonic reducers (Harmonic Drive) a core component for high-end humanoids (like early Japanese series) due to their compactness, high reduction ratio, and nearly zero backlash.
-
Challenges and Evolution: The combination of motors and reducers faces issues of increased volume, weight, and flexibility/impact (the rigidity of reducers leads to poor force relief during collisions). Solutions include:
-
The Path of Hydraulic Drive Power:
-
Highly Integrated: Atlas uses specially designed high-integration valves and actuators to reduce weight and volume.
-
Electric Hydraulic Solutions: Attempting to address noise and oil line issues (like integrating pumps near joints, miniaturization), still in research/prototype stages.
-
Hydraulic Power Units (HPU): Require carrying high-pressure oil pumps (usually motor-driven), oil tanks, cooling systems, and complex piping valve blocks, leading to extremely complex, heavy systems, with significant noise, oil leakage, and maintenance difficulties.
-
Low Energy Efficiency: Pumping and hydraulic circuit leakage losses are significant, far below motor efficiency.
-
Control Precision Challenges: Valve response, fluid compressibility, and piping characteristics affect control precision.
-
Core Principle: Utilizing the incompressibility of liquids to transmit power, adjusting the movement and output force of actuators (cylinders, hydraulic motors) by controlling the opening of high-pressure oil line valves. The advantages lie in extremely high power density and force density, naturally possessing low impedance (can yield passively) and excellent impact resistance (liquids can buffer).
-
Representative Applications: Boston Dynamics’ Petman (early human testing dummy) and the Atlas series (up to the 2023 version) are its peaks. With hydraulic drive, Atlas achieves incredible explosive movements like jumps and backflips.
-
Fatal Drawbacks:
-
Evolution and Exploration:
-
The Dawn and Challenges of Artificial Muscles:
-
Pneumatic Artificial Muscles (PAM / McKibben Muscle): Simple structure (rubber tube with woven sleeve), generating force through inflation and contraction. Advantages: extremely high force-to-weight ratio, good compliance. Disadvantages: low energy efficiency (air pump), control lag (compressed air response), poor position accuracy, requiring complex air source systems. Early research (like Honda’s E0 leg exploration) and a few special platforms (like rehabilitation exoskeletons) used this.
-
Dielectric Elastomer Actuators (DEA): Deforming elastomer materials using electric fields. Fast response, theoretically high efficiency, good compliance. However, output force/displacement is limited, requiring kilovolt-level voltage, and materials are prone to breakdown failure, mostly for laboratory research.
-
Shape Memory Alloys (SMA): Generating force/displacement by recovering preset shapes when heated. Advantages: simple and compact structure, silent. Disadvantages: extremely low efficiency (most energy converted to heat), slow response (heating time), small deformation rate, poor control precision, limited cycle life. Applications are limited to small special movements (like micro-movements, biomimetic insects).
-
Tendon Drives: Drawing inspiration from biological tendons, using high-strength flexible ropes to transmit motor or hydraulic force. Advantages: achieving distal drive (reducing weight at distal joints), designable compliance (like designing low-impedance fingers). Disadvantages include rope friction, elasticity, and hysteresis affecting control precision, requiring complex tensioning and maintenance (especially when coupled with multiple degrees of freedom). Widely used in dexterous hands and humanoid robot arms, waists, and even legs (like Walk-man).
-
Ideal Goals: Simulating the compliance, high power-to-weight ratio, direct linear drive (force/displacement), and natural compliance of biological muscles.
-
Main Types and Explorations:
-
Status and Future: Currently, no artificial muscle technology can comprehensively surpass top electric motor drives and highly engineered hydraulic drives in overall performance (power-to-weight ratio, efficiency, controllability, reliability) for complete humanoids. They are more used for specific components (like fingers) or as cutting-edge exploratory directions. With breakthroughs in material science (new smart materials) and manufacturing processes, disruptive actuators may emerge in the future.
The logic of driving evolution: From purely mechanical to integrated electric/hydraulic drives, from bulky peripherals to highly integrated joint modules, from pursuing single performance parameters (like force) to comprehensively considering power density, efficiency, response bandwidth, impedance characteristics, lifespan, cost, and controllability. High-performance electric drives remain the mainstream choice for practical and large-scale applications (Optimus, NAO, Pepper, latest Electric Atlas), while hydraulic drives, despite showcasing extreme performance (historical versions of Atlas), face significant maintenance challenges, and artificial muscles are still in basic research and specific application development.
2.4 Key Breakthroughs and Turning Points — Defining Directions at the Crossroads
The evolution of humanoid robots is not linear and smooth, but driven by a series of breakthrough achievements and decision events that profoundly influence technological paths, research focuses, and industry landscapes:
-
Key Breakthroughs (Technical Breakthroughs):
-
ZMP (Zero Moment Point) Theory (1968, Miomir Vukobratović): Established a strict mathematical criterion and control theoretical foundation for the dynamic stability of bipedal walking robots, serving as the theoretical cornerstone for transitioning from static to dynamic walking. Its influence is profound (core of early ASIMO).
-
Distributed Force Control and Whole-Body Coordination Control: Early control focused on individual joints/legs, while whole-body coordination control strategies (like task priority and optimization) achieved global coordination of body posture, arm operations, and foot contact forces, which is key to enhancing operational capabilities.
-
Successful Application of Model Predictive Control (MPC) in Walking: Able to calculate the optimal action sequence in real-time over future time windows (considering dynamic constraints and environmental feedback), significantly enhancing the robustness and efficiency of dynamic walking on complex terrains (one of Atlas’s core breakthroughs).
-
Mature High Power Density, High Precision Motor Drivers (like Harmonic Reducers + Brushless Motors): Provided the physical possibility for stable and efficient walking (a hallmark of the ASIMO era).
-
State Estimation and Environmental Modeling Based on Multi-Sensor Fusion (vision, IMU, joint forces/torques): Technologies like SLAM (Simultaneous Localization and Mapping) made autonomous navigation of humanoid robots in unstructured environments possible.
-
Progress in Deep Reinforcement Learning (Deep Reinforcement Learning) in Sim2Real: Enabled robots to learn complex movement and operation strategies through extensive virtual trial-and-error, successfully transferring to physical hardware (like agile running, fall recovery, specific operations), greatly accelerating skill acquisition (representative work from research institutions like DeepMind, ETH Zurich, Berkeley, also supporting the path of Optimus).
-
Key Turning Points:
-
Honda’s Launch of the P Series and ASIMO (1993-2000): For the first time, brought highly dynamically stable bipedal walking technology to public attention (especially the high-profile demonstrations of ASIMO), setting a new benchmark for humanoid robot performance and attracting global attention and imitation, establishing Japan’s early leadership in this field.
-
The Openness of HRP-2 and the Establishment of the OpenHRP Ecosystem (2002): Significantly lowered the entry barrier for research institutions, activating the global academic research community, promoting algorithm innovation and sharing (rather than just closed research by large companies), accelerating technological diffusion.
-
The DARPA Robotics Challenge (DRC) (2013-2015): Designed a series of complex tasks against the backdrop of disaster response, focusing for the first time on the comprehensive operational, mobility, and decision-making capabilities of humanoid robots in real, complex, unstructured environments (rather than just walking). Although the participating robots (mainly improved Atlas, SCHAFT, modified HRP-2, etc.) performed clumsily in the competition, it thoroughly exposed the limitations and vulnerabilities of the technology at the time (especially in operational reliability, environmental perception, human-robot interfaces, and system stability), but also clearly indicated the real pain points and technological priorities in practical applications (autonomy, robustness, fault recovery), becoming an important compass for subsequent R&D.
-
Boston Dynamics Released New Atlas Backflip Video (2016-2021): The ultra-high dynamic movement capabilities showcased shocked the world, redefining people’s perceptions of the physical performance limits of humanoid robots, greatly boosting public and industry expectations for technological potential, while pushing the limits of hydraulic drive performance to new heights.
-
Tesla’s High-Profile Launch and Continuous Promotion of the Optimus Project (2021-Present): Technology giants targeting large-scale commercialization (claiming a cost of $20k) entered the humanoid robot field (especially non-traditional paths based on pure vision + end-to-end AI), bringing massive funding, engineering capabilities, and strong expectations for industry integration. This move attracted more capital and companies to enter (like Figure, 1X, Apptronik, etc., startups receiving significant funding), marking a new phase in humanoid robot R&D shifting from being primarily driven by government/research funding (DRC) to being driven by commercial capital and market expectations.
-
Boston Dynamics Announced Atlas’s Shift to Pure Electric Drive (2024): Marking the engineering community’s choice of the mainstream technological path for the long-term development of humanoid robots — hydraulic drive (though extreme in performance) faces challenges in complexity, maintenance, and noise for commercialization, while highly integrated modular electric drives are more practical and commercially viable under comprehensive considerations. This is a realistic declaration of the logic of technological evolution.
Conclusion:
The evolution of humanoid robots is a legend co-written by dreamers’ blueprints and engineers’ welding guns. From Da Vinci’s concepts to Jaquet-Droz’s precision dolls, from WABOT’s first stumbling steps to ASIMO’s graceful running, and to Atlas’s world-shocking backflips and the large-scale wave led by Optimus, we have witnessed the leap of driving technology from clockwork, hydraulic to highly integrated motors, the iteration of control theory from ZMP static stability to MPC dynamic optimization, and the core focus shifting from performance demonstration to practical operation and environmental adaptability.
This path is filled with technical challenges and unfinished business, but its evolution speed has significantly accelerated with the accumulation of computing power, AI, and capital investment. The organization of this chapter is not only to commemorate the contributions of pioneers but also to understand the inherent logic of technological development, grasp the key forces driving change, and future directions. In the following chapters, we will delve into the internal workings of this complex machine — starting with its “skeleton” (structural design) and “muscles” (driving systems), deconstructing its construction principles one by one.