Fig. 4: Intelligent traffic simulation with RP-RC systems.
From: Dynamic machine vision with retinomorphic photomemristor-reservoir computing
a Simulation of a robot and a car equipped with an RP-RC system. Imprint values (Fig. 3c) are used to generate the first three frames of the PMA with hidden memory. After training by the first three frames of a moving robot and a moving car, the RP-RC systems predict future motion trajectories for both objects based on a single vision input. The first 24 steps of the predicted motions are demonstrated. xcar and xrobot indicate the distance between the car and the robot to the crosswalk. W and L are the width and length of the crosswalk. b Structure of the convolutional autoencoder (CAE) used in the RP-RC systems. c Motion prediction by the RP-RC system for the car. The car will slow down at t = 0 if the robot is predicted to be on the crosswalk at t = xcar/vcar. Otherwise, the car will continue its motion at the same speed. d Motion prediction by the RP-RC system for the robot. The robot will slow down at t = 0 if the car is predicted to be on the crosswalk at t = xrobot/vrobot. Otherwise, the robot will continue its motion at the same speed. e, f RP-RC system decision maps of the car and the robot. Orange and blue areas indicate that the car and robot need to slow down or can keep their speed. g, h Two examples of dynamic decision-making by the car and robot at different initial positions based on motion recognition and prediction.
