Research On Motion Control Of Biomechanics-based Virtual Human

Biomechanics-based human motion research has been widely used in medical rehabilitation,robot motion control computer animation generation and other fields.Our research focuses on the biomechanical virtual human motion control modeling,control system and optimization method,and simulates the generation of stable and realistic virtual human motion in physical environment.Relevant research has important theoretical and application value.The research contents are summarized as follows:1.The virtual human musculoskeletal model based on physiological anatomy knowledge was constructed,which consisted of 7 limbs and 6 joints with 8 internal degrees of freedom,and attach 9 muscle units to each leg.A dynamic model for driving virtual human motion is constructed based on the Hill muscle force model.The muscle tendon units of the musculoskeletal model simulate the mechanical properties of different muscles by setting corresponding internal parameters.The relationship between joint torque and muscle force and the nonlinear variation of muscle length for a given skeletal muscle adhesion relationship in a biomechanical driven virtual human model are clarified.The dynamic model of contact between virtual human and ground is given.All of this provides a theoretical model for the motion control of virtual human in a typical physical environment.2.A multi-layer biomechanical virtual human motion control framework including a policy control layer,a spinal reflex layer,and a muscle driving layer was designed by simulating human motion mechanism.The policy control layer gives the target step size and maintains the body balance by specifying the target angle of the swinging leg so that the virtual human can walk stably at a given speed.The spinal reflex layer can be viewed as a feedback network that receives sensory information from four kind of sensors and maps all inputs to the excitation signals of the muscle tendon units.In the muscle actuate layer,the muscle tendon units generate muscle forces to actuate joints according to the received excitations.A Simulink simulation platform was built for the proposed control system.The biomechanical virtual human motion simulation is realized.3.A stage particle swarm optimization algorithm based on muscle energy consumption and gait similarity is proposed to optimize the control system parameters.A method for extracting key points of motion data is proposed for the evaluation of gait similarity to enhance the realistic of the generated motion.Four simulation experiments were designed and the simulation results demonstrate the effectiveness of the neural reflex control system with particle swarm optimization.The comparison of the results of the first three simulation experiments shows that the proposed optimization method has better motion realistic.The fourth simulation experiment verified the optimization method to improve the ability of the control system to resist external interference.4.Constructing biomechanical virtual human motion control system for environment interaction based on deep reinforcement learning.A deep reinforcement learning strategy algorithm structure is designed for biomechanical driven virtual human continuous motion control.The structure of deep neural network and the observations of virtual human motion state and terrain are designed.There are three main types of obstacle terrain: step terrain,slope terrain and wavy terrain.The use of deep reinforcement learning improves the robustness of the biomechanical virtual human control system to mixed obstacle terrain.