Management of uncertainty using neural networks

The Dempster-Shafer (D-S) theory is one of the methods used to manage uncertainty. However, the main problem concerning the D-S technique is that it suffers from an exponential computational complexity. The paper shows how to implement the D-S combination rule using neural networks to alleviate this problem in the context of a robot safety system. The response time of the neural network measured on a Pentium III 500MHz processor is 1.3ms, well within the required 20 ms standard for robot safety systems used in manufacturing environments. Although the neural network generates small errors, the speed with which it can synthesize sensor data more than compensates for occasional inaccurate responses. The response time of the neural network appears to be independent of the number of sensors used.