Dexterity is one of the most remarkable features of human movement. How to improve the motion ability of robots to realize human-like dexterous and compliant manipulation has long been a core problem in the field of robotics. Compared with traditional articulated robots, the human motor system that consists of muscle, tendon and skeleton, can not only produce explosive force for fast movement but also ensure the stable and accurate force in elaborate operation. On the one hand, due to the highly redundant joints and muscles, the human motor system can complete complex and flexible movements under the modulation of well-developed neural systems. On the other hand, the rigid-flexible coupling characteristics of the musculoskeletal system guarantee the passive flexibility of human body to adapt to variations of environments, so that the favorable factors in the environment are able to be leveraged to assist in completing difficult operations. In addition, the stiffness of the system can be regulated through the synergistic effect between muscles to ensure that the manipulation tasks can be completed accurately. Therefore, based on the structural characteristics and control mechanism of the human motor system, the development of bio-inspired musculoskeletal robots has great scientific significance for further improving the dexterous and compliant ability of robots.
In the past few years, the rapid developments of brain science, neuroscience, artificial intelligence and robotics have provided an important foundation for the interdisciplinary research on brain-inspired intelligent robotics. However, it is still a great challenge to integrate biological mechanisms into robotics. For this interdisciplinary research field, two main bottleneck problems are how to summarize the core mechanisms that are beneficial to improving robot performance from massive neural mechanisms, and how to build the computational models of mesoscopic neural mechanisms to form realizable software and hardware systems.
Based on the research results of brain science and neuroscience, this thesis focus on the dexterous structure analysis and motion control optimization of bio-inspired musculoskeletal robots. By referring to the dexterous structural characteristics of the human musculoskeletal system and neural mechanisms of the brain in movement coordination, we explored the principles of the human motor system in realizing accurate and flexible movement with highly redundant and coupling structure. New kinds of hardware platforms and control methods of musculoskeletal robots are developed in this thesis by mimicking the structure of the musculoskeletal system and neural mechanisms of the brain and spinal cord, which may bring in new inspiration for promoting the dexterity of the robot. The key innovations are listed as follows:
1. According to the biomechanical characteristics of human musculoskeletal system, a quantitative method for computing muscle function is established, and a structure optimization algorithm for the redundancy of the musculoskeletal system is proposed. With highly redundant joints and actuators, the musculoskeletal system of humans can fulfill movement and operation with high precision, flexibility and reliability. However, the redundancy of musculoskeletal system also brings great difficulties to the control and fabrication of human-like robots. Therefore, a structure optimization algorithm is proposed based on convex hull theory to effectively reduce the number of redundant muscles in this work. The dynamics model of musculoskeletal system is first constructed based on the biomechanical characteristics. Then, taking the strength limitation and state-dependent characteristics of muscle into consideration, a feature model for quantitatively described the muscle function is established. By applying the proposed structure optimization algorithm, the muscles corresponding to the vertexes of convex hull formed by the muscle feature vectors are reserved while the redundant muscles within the convex hull are deleted, and a simplified model with high-precision motion ability and less redundant muscles is obtained. Based on the simplified model, a hardware platform of robotic arm has been constructed, which consists of 6 degrees of freedom and 11 bionic pneumatic muscles.
2. Inspired by the neuronal gain modulation mechanism in motor cortex of the brain, a bio-inspired gain-modulated recurrent neural network is proposed to control a highly redundant, coupled and nonlinear musculoskeletal robot. Motor cortex can arouse abundant transient responses to generate complex movements with regulations of neuromodulators, while its architecture remains unchanged. This characteristic endows humans with flexible and robust abilities in adapting dynamic environments, which is exactly the bottleneck problem in the control of complex robots. In this work, inspired by the mechanisms of motor cortex in encoding information and modulating motor commands, a biologically plausible gain-modulated recurrent neural network is proposed, and a novel learning rule that mimics the mechanism of neuromodulators in regulating the learning process of the brain is put forward to train the network effectively. In addition, a new optimization algorithm inspired by the error-based movement correction mechanism in the cerebellum is proposed to control novel movements. Experiments are conducted on an upper extremity musculoskeletal model and a general articulated robot to perform goal-directed tasks. The results indicate that the gain-modulated neural network can effectively control a complex robot to complete various movements with high accuracy, and the proposed algorithms make it possible to realize fast generalization and incremental learning ability.
3. Inspired by the synergistic regulation mechanism of muscle by interneurons of spinal cord, a novel method for constructing constraint force field by optimizing the distribution of muscles is proposed, which can help the musculoskeletal robot realizing high-precision motion with low-precision control. In general, sophisticated structures, highly accurate sensors and well-designed control are all necessary for a traditional robot to achieve high-precision movement. In this work, inspired by two kinds of natural constraints, convergent force field of muscle found in neuroscience and attractive region in environment found in information science, we proposed a structure transforming optimization algorithm for constructing constraint force field in musculoskeletal robots. For the musculoskeletal robot with rigid-flexible coupling and variable structures, constrain force field with strong robustness to noise perturbations can be constructed in the task space of the musculoskeletal robot by optimizing the distribution of muscles. With the help of the constraint force field, the robot can complete precise and robust movement with constant control signals, which brings in the possibility to reduce the requirement of sensing feedback during the motion control of the robot. Based on the formation principle of the constraint force field, a new kind of musculoskeletal robot with variable structure is designed. This work provides a theoretical basis for realizing high-precision manipulation of robots with limited control and sensory accuracy.
|Keyword||肌肉骨骼机器人 生物启发式算法 机器人灵巧运动 环境吸引域|
|Sub direction classification||智能机器人|
|钟汕林. 生物启发式肌肉骨骼机器人灵巧结构与控制研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2021.|
|Files in This Item:|
|Recommend this item|
|Export to Endnote|
|Similar articles in Google Scholar|
|Similar articles in Baidu academic|
|Similar articles in Bing Scholar|
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.