Training an Animated Arm to Reach for Objects in Space Using Neural Networks
In order to analyze the feasibility of using neural networks for automatic motion control, the simple motion of an arm reaching for an object in a two dimensional world was used. Initially, a multi-layered feed-forward network was trained to predict the position of an arm at the following time frame, given only the position of the arm at the current time frame and its relative position from the object. After the network was trained using the back-propagation learning algorithm, the performance of the network was tested, producing an animated motion that was far from acceptable, even though the network's training performance leveled off to a low mean error. It was also discovered that a highly acceptable animated motion could be produced when the network continued to be trained during the testing phase.
17 pages matching determine in this book
Results 1-3 of 17
What people are saying - Write a review
We haven't found any reviews in the usual places.
The Initial Experiments
2 other sections not shown
100'th point plotted 20 iterations 4000 Training Instance 5000 training samples arm configuration arm motion network back-propagation learning algorithm boolean function computer animation constraints current hand position curve for network degrees of freedom desired arm motion desired motion determine elbow angle feed-forward network Figure fixed point attractors forward kinematics goal positions gradient descent hidden units Horn's analytic solution Instance every 100'th inverse kinematics network inverse kinematics solution joint angles keyframes learn the desired Learning curve learning rate mapping mean errors produced mental model momentum network configuration network contains network model network trained network was trained network's performance neural networks Number of hidden number of seconds output units particular training sample path network radius rate was set robot Rumelhart shoulder angle shows the learning sigmoid function simulation specified number straight line path Taylor series testing total input train the network trained to learn training a network user effort weight space