The Next Step in Robot Mobility: Redefining Humanoid Actuators
In the world of advanced robotics, the pursuit of human-like agility and efficiency has long been a driving force behind innovation. Now, a comprehensive new review from researchers at Beijing University of Technology and Shenzhen Ubtech Robotics Co., Ltd. is shedding light on the critical component that could unlock the next generation of bipedal robots: the actuator. Published in the Journal of Harbin Engineering University, the study by Ding Hongyu, Shi Zhaoyao, Yue Huijun, Yu Bo, and Zhang Pan offers a detailed analysis of the evolution, current state, and future trajectory of humanoid robot actuators, positioning the field at a pivotal crossroads.
For decades, the dream of creating robots that move with the grace and power of living beings has captivated engineers and scientists. The iconic image of a robot walking, running, or even leaping with human-like fluidity is no longer confined to science fiction. Yet, despite remarkable progress—from Honda’s ASIMO to Boston Dynamics’ ATLAS—these machines still fall short of the dynamic performance seen in humans and animals. The reason, as this new research highlights, lies not in artificial intelligence or sensor arrays, but in the very muscles that power their movement: the actuators.
An actuator, in the context of humanoid robotics, is far more than a simple motor. It is a sophisticated electromechanical module, integrating a motor, gearbox, encoder, control board, and software into a single joint unit. These components work in concert to generate motion, and their design directly determines a robot’s speed, efficiency, responsiveness, and ability to handle physical impacts. As Ding Hongyu, the lead author and doctoral candidate, explains, “The actuator is the heart of the robot’s locomotion. If we want robots to move like humans, we need actuators that can match the performance of biological muscle.”
The paper traces the lineage of actuator development back to 1971, when Professor Kato Ichiro of Waseda University unveiled WAP-3, the world’s first three-dimensional bipedal robot capable of static walking. This milestone marked the beginning of a new era in robotics, one defined by the challenge of mimicking the complex mechanics of human gait. Early designs relied on what the authors term “traditional stiffness actuators” (TSA), which paired high-ratio gearboxes with electric motors to generate the necessary torque. While effective for basic motion, these rigid systems proved ill-suited for the dynamic, unpredictable environments where robots are increasingly expected to operate.
The limitations of TSA became apparent as researchers pushed the boundaries of robot mobility. A rigid actuator, much like a stiff spring, transmits force directly and instantly. While this allows for precise position control, it also makes the system vulnerable to shock and damage during impacts—such as when a robot’s foot strikes the ground. Moreover, the high gear ratios required to amplify motor torque result in low power density, a key metric that measures how much power an actuator can deliver relative to its weight. Biological muscle, for comparison, achieves a power density of around 500 watts per kilogram. Traditional actuators, even with decades of refinement, struggle to exceed 300 W/kg, creating a significant performance gap.
Recognizing these shortcomings, the field took a major leap forward in 1995 when Gill Pratt and his team at the Massachusetts Institute of Technology introduced the concept of the series elastic actuator (SEA). The innovation was deceptively simple: insert a spring between the motor and the load. This elastic element acts as a mechanical buffer, absorbing impact energy and allowing for smoother, more compliant motion. The benefits were immediate and profound. SEAs demonstrated superior force control, reduced reflected inertia, and the ability to store and release energy—much like tendons in the human body.
The impact of SEA technology has been far-reaching. Robots such as NASA’s Valkyrie and the Italian Institute of Technology’s Walk-Man have adopted elastic actuators to enhance their safety and adaptability in human environments. The design enables them to interact with objects and terrain without causing damage, a critical requirement for collaborative robots. However, the paper notes that the introduction of elasticity comes at a cost. The system becomes “underactuated,” meaning it has fewer control inputs than degrees of freedom, which complicates motion control and reduces precision. This trade-off has spurred further innovation, leading to the development of parallel elastic actuators (PEA), clutched elastic actuators (CEA), and multi-mode elastic actuators (MEA).
Parallel elastic actuators take a different approach, placing the spring in parallel with the motor rather than in series. This configuration allows the spring to share the load, reducing the peak torque required from the motor and improving energy efficiency. Studies cited in the review show that well-designed PEA systems can cut motor torque demands by up to 50% and reduce energy consumption by 25%. When combined with a clutch mechanism, as in CEA designs, the spring can be engaged or disengaged on demand, enabling precise control over energy storage and release. This capability is particularly valuable in tasks that involve repetitive motion, such as walking or running, where energy can be recycled from one step to the next.
Multi-mode actuators represent the cutting edge of this evolution, combining elements of both series and parallel elasticity into a single, highly adaptable system. These actuators can switch between different modes of operation, optimizing performance for specific tasks. For example, a robot might use a stiff mode for precise manipulation and a compliant mode for safe interaction with humans. The research highlights work by Mathijssen and Geeroms, who have developed prototypes that mimic the locking mechanisms found in biological joints, allowing for both high efficiency and dynamic responsiveness.
Despite these advances, the complexity of multi-mode systems presents new challenges. Their mechanical design is inherently more intricate, and their control algorithms require sophisticated modeling to manage the interactions between multiple elastic elements. As a result, widespread adoption in commercial robotics has been limited. This is where the third major category of actuators—proprioceptive actuators, or “quasi-direct-drive” (PA) systems—comes into play.
Introduced in 2016 by Wensing and colleagues, PA systems represent a radical departure from both TSA and SEA designs. Instead of relying on high-ratio gearboxes or external springs, they use high-torque-density motors paired with low-ratio gearboxes. The result is a compact, lightweight actuator with exceptional power density and bandwidth. The MIT Cheetah robot, which can run at speeds of over 10 miles per hour, is powered by PA units that achieve a staggering 1,416 W/kg—nearly three times the power density of biological muscle.
The key innovation in PA systems is their ability to perform force control without the need for external force or torque sensors. By precisely monitoring the motor’s current and position, the controller can infer the interaction forces between the robot and its environment—a capability known as proprioception. This not only reduces system complexity and cost but also enables faster, more responsive control. As Katz, one of the developers of the improved PA design used in the Cheetah and Hermes robots, notes, “The ability to sense force through the motor itself opens up new possibilities for dynamic locomotion.”
However, even PA systems are not without their challenges. One persistent issue is the problem of homing—how to determine the absolute position of a joint after a power loss. Most PA designs use a single encoder on the motor side, which means the system loses track of the joint’s position when powered down. While solutions such as dual-encoder setups and battery-backed multi-turn encoders have been proposed, they add complexity and cost. The authors emphasize that resolving this issue is critical for the broader adoption of PA technology in humanoid robots.
The review goes beyond a mere catalog of technologies, offering a clear-eyed assessment of the current state of the field and a roadmap for future research. One of the most compelling insights is the growing importance of bio-inspired design. Rather than treating the actuator as an isolated component, the authors advocate for a holistic approach that integrates the actuator with the robot’s overall structure and control system. By studying the musculoskeletal systems of animals—such as ostriches, quails, and other bipeds—engineers can gain valuable insights into how to optimize joint configuration, energy storage, and dynamic stability.
This systems-level thinking is already yielding results. Researchers at the University of Michigan, for example, have developed a robotic leg that mimics the spring-like behavior of avian tendons, achieving remarkable energy efficiency during running. Similarly, work at ETH Zurich on the ANYmal quadruped robot has demonstrated the benefits of combining elastic actuators with advanced gait planning algorithms to navigate rough terrain.
The authors also highlight the need for standardization in actuator testing and performance evaluation. While the field has made significant progress, the lack of common metrics makes it difficult to compare different designs and assess their real-world performance. Establishing standardized benchmarks for parameters such as backlash, starting torque, stiffness, and efficiency would accelerate innovation by providing a clear framework for comparison and improvement.
Looking ahead, the paper identifies several key trends that will shape the future of humanoid actuators. In the realm of rigid actuators, the focus will shift toward refining existing designs and establishing performance standards, particularly for industrial applications. For elastic actuators, the challenge will be to integrate them more fully into the robot’s overall architecture, optimizing not just the actuator itself but the entire system for maximum efficiency and performance. And for quasi-direct-drive systems, the primary goals will be to improve encoder technology and further increase motor power density.
Another emerging trend is the evolution of communication protocols. As robots become more complex and interconnected, the need for efficient, reliable communication between actuators and central controllers grows. The authors suggest that wireless communication, enabled by 5G and cloud technologies, could eliminate the need for internal wiring, reducing weight and improving reliability. Real-time monitoring of actuator health and performance could also enable predictive maintenance and adaptive control strategies.
The implications of this research extend far beyond the laboratory. As humanoid robots become more capable and affordable, they are poised to play an increasingly important role in society. In homes, they could assist the elderly and disabled with daily tasks. In disaster zones, they could enter environments too dangerous for humans. In factories, they could work alongside people, performing tasks that require both strength and dexterity.
But for these visions to become reality, the actuators that power them must continue to evolve. The review by Ding Hongyu and his colleagues provides a crucial foundation for that evolution, offering both a historical perspective and a forward-looking analysis. It underscores the fact that the future of robotics is not just about smarter algorithms or better sensors, but about rethinking the very mechanics of movement.
As the field moves forward, the collaboration between academia and industry will be essential. Institutions like Beijing University of Technology and companies like Ubtech are already working together to bridge the gap between research and commercialization. With continued investment and innovation, the day when robots move with the same grace and power as humans may not be as far off as it once seemed.
The journey from WAP-3 to the next generation of humanoid robots has been long and complex. But with each new actuator design, each breakthrough in materials and control, the dream of truly human-like robots comes a little closer to reality. The work of Ding Hongyu, Shi Zhaoyao, Yue Huijun, Yu Bo, and Zhang Pan is not just a review of the past—it is a blueprint for the future.
Ding Hongyu, Shi Zhaoyao, Yue Huijun, Yu Bo, Zhang Pan, Beijing University of Technology and Shenzhen Ubtech Robotics Co., Ltd., Journal of Harbin Engineering University, DOI: 10.11990/jheu.202003028