Haptic communication system using cutaneous actuators for simulation of continuous human touch
摘要
說(shuō)明書
The neural network 1310, during training, is also constrained based on the various constraints 1312, which will be described in further detail below. These constraints 1312 aim to constrain the output of the neural network 1310 towards high compressibility, low entropy, low ordinality (i.e., fewer number of possible haptic outputs), sparsity, temporal stability, and spatial stability. Subsequent to the training process, the neural network 1310 is able to produce a set of haptic cues for a set of cutaneous actuators which represents the input acoustic signal 1302. When the haptic outputs indicated by these haptic cues are sensed by a user, the user, who may be trained, may then be able to recreate, or at least have an understanding of the input acoustic signal. The haptic output, if recorded, may also be used to reconstruct the original input acoustic signal. In this fashion, an acoustic signal can be compressed into a haptic (or tactile) signal.
The cost analyzer 1316 determines, during the training of the neural network 1310, the cost of the current iteration of the neural network 1310. The cost is computed based on the error between a reconstructed version of the acoustic signal using the current output of the neural network 1310 in addition to how well the current iteration of the neural network 1310 meets the constraints 1312.
