Research On Adaptive Neurobehavioral Control For Legged Robot

Walking animals have shown energy efficiency and versatile locomotion ability when adapting to their environments,and different methods,including machine learning,model control and biologically inspired control,have been applied in artificial legged locomotion systems to imitate these natural characteristics.But most of them can only imitate the simple behavior of animals in nature and make less appropriate response to the changes of environmental factors,which is far from their natural characteristics.In order to improve the adaptability and behavior diversity for legged robots,an adaptive and diverse behavior control strategy is proposed,which is used to integrate multiple neural modules to build neural behavior control architecture for robots.According to different topographic features and environmental conditions,this neural architecture can realize rapid switching of behavior mutations and form various complex movement behavior for robots.The overall system mainly consists of adaptive neuromodulation module and motor neurocontrol module.The function of adaptive neuromodulation module is to process the information of the external environment and convert goal-directed behavioral instructions into nerve signal strength,which can obtain animal-like signal processing mechanisms and behavioral responses,and achieve the adaptive adjustment of body and legs of the robot.This module includes the signal preprocessing neural module and the neural strength modulation module.The signal preprocessing neural module is used to deal with sensing signal and convert it into behavioral instructions.After receiving the behavior instructions,the neural modulation module converts it into neural signal strength,which can provide precise real-time control of all variables of robot motion behavior,including attitude and trajectory of the body and the state and trajectory of foot-tip of robot,enabling more dexterous and fast movement behavior.The agile motor neurocontrol module is used to generate the basic motion behavior of the body and legs of the robot,including the motion coordination neural modulation module and the virtual motoneuron module.Motion coordination neural modulation module mainly generates rhythm and gait signal to maintain the coordination between legs.The proposed virtual motoneuron module forms different body and foot-tip trajectories by receiving the rhythm signals from the motion coordination neural modulation module and strength signals from nerve strength modulation module,and then the signals driving the robot joint motor is obtained through the inverse kinematics module.In addition,the virtual motoneuron module has the mathematical characteristics of Bessel curve and the adjustable parameters and layers of neural network.As a behavior planning module of legged robot,it can adjust the complexity,accuracy and controllability of behavior trajectory according to the changes of environmental parameters.At the same time,combining the neural network-based robotic upper-layer adaptive behavior control network with the virtual motoneuron module can complete the real-time adjustment of all the motion behaviors of all parts of the robot.Finally,the diversity motion experiments,the body adaptive obstacle avoidance experiments,the leg adaptive obstacle avoidance experiments and the robot versatility verification experiments of the robot are carried out on a hexapod robot experimental platform.Experimental results show that the neural architecture proposed in this paper can automatically adjust the trajectory planning neural module according to the environmental changes,so that the robot can flexibly change the motion form through various complex environments.This strategy allows the robot to respond flexibly and steadily to environmental changes,demonstrating good self-adaptability.