Autonomous agents plan their paths through known and unknown environments to reach their goals. When mu ltiple autonomous agents share the same area, conflict situations may occur that need to be solved. We present a decentralized decision mak ing algorithm to solve conflicts among autonomous agents. It is based on two main ideas: First, we in troduce an innovative operationalization of cooperative behavior which allows to determine whether a behavior is cooperative by computing the total utility and com paring it to a reference utility. Second, we use motion primitives as a representation of available maneuvers obeying individual and environmental restrictions. The decentralized decision making algorithm is based on communication among the autonomous agents to find an optimal maneuver combin ation. Simulations show that our algorithm is applicable to different highway traffic scenarios of two automated vehicles. We use a mean-square acceleration as an individual cost function and show that our intelligent controller leads to coop erative solutions.

No abstract

No abstract

Agents can benefit from cooperative behavior as it intends to increase the total utility. In traffic situations, one agent may behave cooperatively by yielding to another. In recurring situations, this may lead to an imbalanced distribution of the benefit. However, humans prefer balanced utility distributions over imbalanced ones. While state-of-the-art cooperative decentralized decision making may promote such an imbalanced utility distribution, advanced cooperative decentralized decision making can support equality among the agents. Three different approaches are compared: Considering the agents' utilities perfect substitutes, imperfect substitutes, and perfect substitutes with time-variable rates of substitution based on a cooperative reward system. Simulations of a highway scenario reveal the differences in recurring situations: Perfect substitutes indeed maximize total utility, but at the expense of a highly unequal utility distribution that may lead to poor long-term user acceptance; imperfect substitutes promote an equal utility distribution, but leave much of the potential of cooperative behavior unused; introducing a cooperative reward system based on memories of costs is shown to allow for a trade-off between both-altruistic-cooperative behavior without constant preference of one agent over the other, which is assumed to improve user acceptance.

No abstract