The green hands aim at a target at the top of the screen, while avoiding the black obstacles. The actual state of the system is uncertain, and the eye provides local disambiguation of the "real" state.
I also model visual inhibition during saccades - the disambiguation is inversely proportional to the eye's velocity, so the eye provides no useful information when it is in fast motion.
The system is formulated as a stochastic optimal control problem, solved using differential dynamic programming over an augmented state space (representing the distribution in the real state space as a single Gaussian, estimated using the Kalman filter).
Link to this comment:
All Comments (0)