Search

PETREL, a winged hybrid-driven underwater glider is presented, which features long-endurance low-noise while penetrating the ocean like a UW glider and burst-speed high-maneuverability while operated as an AUV. In addition, kinematic and dynamic model of the vehicle are developed based the theorem of momentum and moment of momentum. Moreover, equations of equilibrium indicating the balance pitch and attack angles are deduced and indexes for evaluation of maneuverability of this vehicle in 6-DOF are defined. Furthermore, its specific characters are investigated by simulation and the functionality is validated by field trial. In thrust operation, the external bladder for buoyancy change improves the efficiency of pitch control while mounted on the fore end of vehicle and the swerving radius is largely decreased by using vertical fins for steering instead of roll mass; Meanwhile in glide operation, the longitudinal maneuverability is largely increased due to the fixed wings and the capability of attitude keeping is augmented by redistribution of internal mass (roll and pitch mass) while the vehicle cruises at low speed.

Depth parking control is important for the HTPE (HydroThermal Plume Explorer). It aims to have a more accurate parking depth and a less oscillation at the target depth. The plant approaching and motion forecasting model based on RBF neural network is built to self-adjust and approach online motion law of HTPE in ideal circumstance of the sea, and simulate output of given control quantity after an PID control period. Together with RBF neural network model, BP neural network based self-adaptive PID control model is used for rectification of PID parameters and for holding the parking depth. Finally, simulations testify that neural networks based control method is feasible.

A new method for smooth trajectory planning of a 6-DOF robot in joint space is described in this paper. By the researching processes of concrete analysis of trajectory planning on robot's manipulator arm, imitation of trajectory based on kinematics and optimization of trajectory in the joint space, an one-input-six-output RBF neural network model is built and trained taking the discrete time as input and the values of six angles as outputs in joint space. With character of rapid convergence and near approximation, this new algorithm is fault tolerant and irrelative with order of inputs, which can ensure the result trajectory is firing enough. The algorithm has been tested in simulation when the virtual model of the robot was established in software ADAMS, yielding good results by studying the kinematics and the dynamics performance of the robot.