Neural network architecture for robot hand control

Connectionist approaches to the control of manipulation robots at the executive hierarchical level: An overview

“…Also, inverse dynamic mapping is more complex, because dependence on contact forces is included. There are a few interesting connectionist solutions in this area [70][71][72][73][74].…”

An experimental approach to robotic grasping using a connectionist architecture and generic grasping functions

“…This learning algorithm is a gradient descent algorithm that is designed to minimize a cost function, which is the mean square error between the desired output of the network and the actual output. Thus (13) Using the chain rule and differentiating, the final learning rule becomes (14) where is a learning coefficient and is defined as (assuming a sigmoid transfer function) for each of the output layer processing elements, where is the actual output of the th PE and is the desired output.…”

“…On the other hand, several researchers have proposed the use of grasping rules that are based on studies of how humans grasp and manipulate objects [1], [3], [8], [14], [22]. These rules, which are often based on sensory input, allow more irregular or "ecological" objects to be grasped.…”

Artificial Neural Networks in Robotic Applications

“…The other papers on the applications are; the feedback loop and the inverse dynamics model for learning trajectory control of an industrial robotic manipulator [40], for general robotic control and a computed torque controller [64,65], one dimensional pole-balancing problem [19] by reinforcement learning [66], the control of the tracking behaviour of an autonomous mobile robot [67,68], three different neural architectures examined by [45] for nonlinear robotic control, the inverse Jacobian control with a hierarchical neural networks [26,69], the structured hierarchical neural network [59] for the real-time 'control of mobile robots, the design and control multi-degree-of-freedom articulated robot han@s [50,[70][71][72][73][74], generating suitable grasp modes for various robot hands and for various objects [72], a gripper with three straight fingers [73], control of a robot arm and gripper for the task of grasping a cylinder [71], a recentralised variable structure control system [74], servo controller [70] for one or two dimensional robotic manipulators, direct transition from/lsensor processing to inverse dynamics [75], controlling the position of robot manipulator [76], guiding the end of the manipulator by CMAC [77], a mobile robot [78} for multi-tasks, underwater robotic vehicles [79][80][81][82] for increasing the autonomy of the vehicle control, recovering the image, teaching the pitch attitude of underwater telerobot with the combination of ANN and fuzzy logic, the neural compensation techniques by [83] for robot control, several ways for an existing controller development, direct adaptive control, optimised non-linear controller development [33,54,84], for time-optimal control …”