| As the third largest country in the world,China has a large area of lake,sea with a long coastline,and web land.There is a lot of demands for amphibious operations in the areas.Amphibious robots,which can walk freely in various complex terrains,have become effective tools for the amphibious operations.When performing repetitive operations,the real-time tracking of desired trajectories is an challenging problem due to the complex kinematics model of amphibious robots as well as the terrain fluctuation and variable medium in field environments.To solve the problem of trajectory tracking for amphibious robots,the thesis proposes a model predictive control algorithm based on the kinematic model learned with a deep neural network.Amphibious robot motion data from simulation is collected and feed into the deep neural network for training,and the kinematic model of the amphibious robot with a given map is obtained.A terrain elevation model is built by using the point clouds obtained from a depth camera,while a robotic elevation model is built by using the CAD model.Then the two evaluation models are compared in pixel-level to determine the contact point between the robot foot and the terrain.By calculating the error function,the robotic body's normal vector fitted by the contact point is obtained.Finally,the optimal control sequence for real-time trajectory tracking is obtained by solving the cost function constructed based on the proposed DNN-based kinematics model and attitude estimation algorithm.To verify the performance of the proposed approach,a Gazebo simulation platform is built,containing simulated terrain maps and amphibious robot.The gait of amphibious robot is also designed.Several simulations are performed,and the tracking error,average error,maximum error,and safety rate are analyzed.The results show that the proposed trajectory tracking algorithm performs better than the traditional trajectory tracking algorithm in both robustness and accuracy. |
PDF Full Text Request