For us, sensor-based navigation means real-time terrain traversal by simulated human agents [19][32]. Agents begin in arbitrary positions in an outdoor environment. During a simulation they walk to the goal, avoiding obstacles and each other, ducking under low branches, climbing over objects, avoiding difficult terrain where feasible, and squeezing through tight spaces where necessary. In addition, the agents attempt to evade detection by hostile agents by avoiding the sensory fields of any hostiles in the area.
An agent is not told how to reach its goal nor is off-line path-planning used. Instead, it is made aware of its environment through the use of a network of simulated sensors. These sensors acquire information on object size, type, and location, passageways, terrain type, exposure to hostile agents, and so on. Based on this information the path through the terrain is incrementally computed, one step at a time. This allows the agent to react to moving obstacles, changing terrain, or unexpected events due to hostile agents or the effects of limited perception.