Bio-inspired Approaches for Human Locomotion: From Concepts to Applications

Abstract

After millions of years of evolution, humans can achieve locomotion tasks in complex environments with versatile, robust and efficient bipedal gaits. Understanding human locomotion control systems can help us develop novel bio-inspired based methods for improving the current legged robots (e.g. humanoids) and wearable devices (e.g. prostheses, exoskeletons). This thesis systematically explores the bio-inspired approaches from concepts to applications for further understanding human locomotion. It includes three main parts: biomechanical studies on human experiments, hardware implementations of bio-inspired concepts, and modeling of human locomotion. The biomechanical studies provide insights on the human locomotor control systems. Human locomotion control can be separated into three locomotor subfunctions which are stance (axial leg function), swing (rotational leg function), and balance (posture control). We investigated how these subfunctions interact with each other by analyzing the contribution of stance and swing leg movements to the walking dynamics. The results reveal a coupling mechanism and synergistic interactions between the subfunctions. Further analyses on the human gait initiation (from standing to walking) experimental data demonstrate that the swing leg and stance leg functions are emerged during the first stride of the stance limb. And we find a strong correlation between the control of the frontal plane and the sagittal plane joints. All these results indicate that the support of one subfunction can provide benefits for the others. Inspired by the findings from the previous biomechanical studies, we implemented bio-inspired balance control strategies on a lower-limb exoskeleton for human walking. The hardware implementations are used to validate and demonstrate the benefits of bio-inspired control concepts. The results show that the bio-inspired balance controller can not only support the swing and stance leg function but also reduce the metabolic costs and assist human walking. The results also support the prior biomechanical studies which suggest synergistic interactions between the subfunctions. In addition, we also implemented a bio-inspired neuromuscular reflex based controller for a hopping robot to investigate the potential benefits of the muscle properties for the stance (rebounding) leg function. The results demonstrate that the robot can achieve stable and robust hopping with the bio-inspired controller. Further analyses show that the neuromuscular properties play an important role in stabilizing the motion. These results indicate that gait models which include the muscle properties and reflex-like control could better reproduce human locomotion. The modeling of human locomotion help us test the bio-inspired concepts in the simulation and reveal the key components of human locomotion control. Here, based on the previous findings, we developed a complex neuromuscular gait model to produce subject specific walking behaviors. Deep reinforcement learning methods were used to generate the control policy (sensor-motor mappings) which has similar functionality as human spinal cord neural circuitries. The results show that the model can achieve robust walking and closely reproduce human joint kinematics and muscle activations. In addition, we also found that the neuromuscular dynamics can facilitate the learning. In future, the proposed gait model can be used to identify optimal control schemes for wearable robots (e.g. prostheses, exoskeletons). In summary, this thesis presents a systematic approach of investigating bio-inspired concepts for human locomotion by experimental studies of human gait, simulations, and hardware implementations. The main contribution of this work is demonstrating how the bio-inspired concepts are extracted from the human experimental data, tested with the simulation models, and implemented and validated with the hardware systems. The outcomes of this thesis can be used as a framework to develop novel bio-inspired controllers for improving the performance of legged robots (e.g. humanoids) and wearable robots (e.g. prostheses, exoskeletons).