Instructions for Interaction with the Environment

There is a point-like agent with ``active perception.'' It can move about in a two-dimensional environment (details of the environment are described in Section 4). The agent's current position is given by a pair of real values. Its current direction is given by variable Direction with range $[0, \ldots, 2 \pi ]$.

MoveAgent(): Move the agent $12$ unit lengths in direction Direction, unless an obstacle or the environment's boundary blocks the path (the value 12 was chosen arbitrarily -- neither it nor any other arbitrarily chosen constants mentioned below have ever been changed during the experiments). Moving may lead to new inputs: $\cal S$$_{0}$, $\cal S$$_{1}$, $\cal S$$_{2}$ become equal to the agent's horizontal position, vertical position, and current direction, respectively. Here position components are rounded to integers between 1 and 1000, and directions are scaled to $[0, 100]$ and then rounded. If the point $24$ unit lengths in front of the agent is within an obstacle or outside the enviroment, then $\cal S$$_{3}$ $\leftarrow$ $24$ (the value 24 is chosen arbitrarily). Otherwise $\cal S$$_{3}$ $\leftarrow 0$. Similarly for $\cal S$$_{4}$ (input 90 degrees to the right), $\cal S$$_{5}$ (behind), $\cal S$$_{6}$ (to the left).

SetDirection($x_1$): Direction $= \frac{x_1}{n} 2 \pi$. Compute new inputs like instruction MoveAgent().

Back to Active Learning - Exploration - Curiosity page