Social potential fields are a way to control autonomous
agents using inverse-power laws on attractive and repulsive forces between the
agents and objects of the environment. For the social potential field agent we
have implemented an agent whose movement is determined by a set of forces which
attract or repulse the agent to agents and object of its sensor field. The
resulting force is:
where 'vij' is the unit vector of the direction
from agent 'i' to agent 'j', and 'r' is the distance from agent 'i' to agent
'j'. The parameters 'c1' ,'c2' >= 0 and 'σ1', 'σ2' > 0 are
determining the nature of the forces between the agent and the object.
Once we decided on the general form of the forces, the next step is the choice
of the parameters 'c1', 'c2', 'σ1' and 'σ2' such that the desired
behavior of the agent is obtained. In practice, the determination of these
parameters is a result of experience and experimentation. We have determined
four sets of these parameters, which describe the relationship of a social
potential field agent to (1) another social potential field agent, (2) an other
agent, (3) food items and (4) obstacles. The experimentally obtained values are
displayed below.
|
|
During testing, two major problems were found with the movements of the agents.
Agents had a tendency to be stuck to into local minima, such as becoming
immobile in the geometrical center of several food sources. Second, agents
frequently overshot the food location and performed an oscillatory movement
around it. A similar problem led to the agent bouncing indefinitely between two
obstacles. These problems were solved by adding heuristics which (a) break the
tie between the attraction forces and (b) prevent repetitive movements.
As the SPF paradigm describes only the movement of an agent, we applied a set
of simple heuristics for the remaining actions. The attack heuristics dictates
that the agent attacks any agent which gets closer than half of the critical
distance. The mating heuristics encourages the
mating of isolated SPF agents, but restricts the mating of SPF agents inside
groups. As an emergent property, this heuristics leads to moderate size,
relatively stable groups of SPF agents.
|
