Michelangelo’s famous painting “The Creation of Adam” captures a moment where God and Adam extend their fingers towards each other, that earth-shattering touch is the moment God bestows life and soul upon humanity.
Image source: Internet
In the world of robots, humans play a role similar to that of God. Robots possess a systematic “body language” that allows them to flexibly perceive complex external environments and efficiently complete tasks. This is no longer a distant fantasy from science fiction, but a technological revolution rapidly advancing in our era. According to Professor Ma Shugen from the field of robotics and autonomous systems, researching robots is key to breakthroughs in the field. So, how can we make robots ‘smart’ like living beings?How can we conduct bionic robot research?What is the future direction of robot development? Today, let’s explore the “body language” of robots and the evolution of robot research!
A fish that has lost its life can still “swim” in water at a certain flow rate. This interaction between body morphology and the external environment is an important manifestation of embodied intelligence.
Image source: Internet
Embodied intelligence includes two types: intrinsic intelligence, which is responsible for maintaining dynamic balance within the body; and acquiring information from the external environment, adjusting internal states based on this information to adapt to the outside world. Long before the concept of embodied intelligence emerged, bionics had already begun developing robots by studying the dynamic walking of living organisms. Researchers use knowledge of biomechanics to study the movement patterns of organisms, employing existing control and perception methods to achieve interaction between robots and their environments. Living organisms evolve different shapes and movement patterns based on their environments; for example, snakes exhibit different movement forms on ice, flat ground, and rugged terrain, which is a manifestation of their intrinsic intelligence. In the past, the field of artificial intelligence focused more on virtual capabilities such as computation and data processing. The emergence of embodied intelligence has renewed the importance of hardware. People are now beginning to think about how to transfer the intelligent aspects of the brain to the body, enabling interaction and operation in the real world. Robots are one such physical entity.
Big data is fundamental. Based on big data, artificial intelligence, such as large language models, uses supercomputers to achieve so-called intelligence. Computational power refers to the computing capability of computers, and the level of intelligence obtained depends on the amount of data and computational power. However, even though large models can help process vast amounts of data and demonstrate great potential in the perception-decision-making process, their “black box” nature raises reliability issues. Mechanism research is key to the research and development of robots. Mechanism refers to the intrinsic principles of things. Every entity has intrinsic principles, which depend on physical properties. For instance, the shape of a river is ultimately determined by the hydrodynamics; the starting forms of a 100-meter sprint and a marathon differ based on the required acceleration at the start. The external appearance of anything has its intrinsic reasons, which are the external manifestations of its internal causes. Research cannot only focus on surface phenomena; it must investigate and discover intrinsic principles (mechanisms). Biological mechanism research is generally conducted by formulating hypotheses and then proving them through experiments. Mastering mechanisms and mechanism models allows for the study of various properties and characteristics without relying on big data and computational power. However, currently, there is no definitive conclusion on the exploration of biological mechanisms, nor is there a unified method to directly verify their correctness. The so-called biological mechanisms still stem from hypotheses that need to be validated through experiments on actual robots.
In the face of complex real-world environments, the closed loop that robots must respond to is: sensor data acquisition — obstacle recognition — action planning — action execution. If any link fails to respond in a timely manner, it can lead to disconnection in the other links, increasing the likelihood of collisions when the robot is in high-speed motion. Based on this, Professor Ma Shugen’s team proposed a concept: not to rely entirely on cognition to complete actions, but to achieve adaptation to the environment through mechanism design. The team has reasonably utilized underactuated systems in mechanism design. For example, when a machine encounters an obstacle, traditional thinking suggests lifting or avoiding it, but they attempt to passively change the movement pattern based on the thrust from the obstacle to continue moving forward. This design approach emphasizes utilizing external environmental factors to achieve machine actions, allowing for problem-solving through structural adjustments, thus realizing a more reliable and efficient mode of movement.
There are two advantages to studying snake-like robots: first, snakes have a simple structure, resembling a rope-like biological form; second, a snake’s small brain can control its entire body movement, which inspires research on how to maximize functionality with limited resources. In terms of application scenarios, snake-like robots can be used in deep-sea exploration, pipeline inspection, and rescue operations in narrow spaces such as collapsed buildings.
Researching bionic robots is not about complete imitation, but rather learning first, then breaking and recreating. For example, humans initially attempted to fly by imitating birds flapping their arms up and down, which failed; later, they invented airplanes using core engines and fixed wings, leading to the emergence of foundational theories such as aerodynamics. Drawing inspiration from nature, we develop robotic machines by imitating organisms and exploring their bionic control strategies, through the process of “guarding, breaking, and leaving”, that is, “learning first, then breaking and recreating” to create new robotic systems.
Research on environmentally adaptive robots that imitate the excellent functions and movements of organisms has not only created various robot mechanisms that can adapt to environments but also developed controllers based on neural oscillators and energy “dissipation, conversion, and balance,” enabling robots to achieve high mobility, low energy consumption, and stability in complex environments.
For bionic robots, a research closed loop should be: imitating organisms to create robots — using robots to understand organisms — improving robots after understanding organisms — and further understanding organisms. This way, humans can better understand nature and create better machines. However, in the field of humanoid robots, the aspect of “breaking” still needs to be discovered. Currently, humanoid robots primarily rely on motors, and technologies simulating joint muscles are still under development, making it unlikely to completely replace humans in the short term.
Increasing global economic output is not the key to the development of the robotics industry; its impact on human social development is what we need to focus on. Research on robots should emphasize their auxiliary functions; they should assist humans in completing dangerous, heavy, or undesirable tasks. Imagine if all high-risk engineering tasks could be automated by robots, allowing humans to be liberated to enjoy life; that would be the ideal application scenario for robots. Rather than the current promotional concepts where robots replace humans in everyday family scenarios, especially in caregiving roles such as elderly care and childcare, where human emotional support is irreplaceable.
Professor Profile
Dr. Ma Shugen
●Professor of Robotics and Autonomous Systems at the Hong Kong University of Science and Technology (Guangzhou)
Dr. Ma Shugen is an honorary professor at Ritsumeikan University, a foreign academician of the Japan Academy of Engineering, a fellow of the Institute of Electrical and Electronics Engineers, a fellow of the Asia-Pacific Artificial Intelligence Society, a fellow of the Japan Society of Mechanical Engineers, and the chair of the prestigious IROS 2022 conference in the robotics field. He has organized and led numerous academic conferences such as IEEE. His research includes bionic robot mechanism design and control, high autonomy robots adaptable to various environments, and amphibious robots. He has published over 530 papers, received more than 30 best paper awards or nominations at international conferences, holds over 50 authorized invention patents, and has signed licensing agreements with several companies for product commercialization. He has supervised 45 doctoral students and over 100 master’s students, including recipients of the Excellent Young Scientists Fund and nominees for national-level research talents. He is listed among the top 2% of the most influential scientists globally by Stanford and on the lifetime scientific influence list.
ROAS
Source: Professor Ma Shugen, AI Technology Review
Editor: Yun Xitong Creative Hub
Source: Professor Ma Shugen, AI Technology Review
Provided by: System Hub
