Opponent Process Control in Linked, Dynamical Agents
نویسندگان
چکیده
Previously, we showed that a minimally cognitive, visual agent demonstrated selective attention and reactive inhibition (Ward & Ward, in press). Surprisingly, we discovered the existence of an opponent-process architecture in our agent’s evolved, neural-network controller. Here, we investigate how opponent processes affect response control in the agent. We scale up cognitive problem solving by evolving links between multiple copies of the visual agent to solve tasks that a single agent cannot work out alone. Opponent processing effects are demonstrated in the linked agent’s response control.
منابع مشابه
Combining Opponent Modeling and Model-Based Reinforcement Learning in a Two-Player Competitive Game
When an opponent with a stationary and stochastic policy is encountered in a twoplayer competitive game, model-free Reinforcement Learning (RL) techniques such as Q-learning and Sarsa(λ) can be used to learn near-optimal counter strategies given enough time. When an agent has learned such counter strategies against multiple diverse opponents, it is not trivial to decide which one to use when a ...
متن کاملRecognizing Probabilistic Opponent Movement Models
In multiagent adversarial domains, team agents should adapt to the environment and opponent. We introduce a model representation as part of a planning process for a simulated soccer domain. The planning is centralized, but the plans are executed in a multi-agent environment, with teammate and opponent agents. Further, we present a recognition algorithm where the model which most closely matches...
متن کاملDynamical Control of Computations Using the Family of Optimal Two-point Methods to Solve Nonlinear Equations
One of the considerable discussions for solving the nonlinear equations is to find the optimal iteration, and to use a proper termination criterion which is able to obtain a high accuracy for the numerical solution. In this paper, for a certain class of the family of optimal two-point methods, we propose a new scheme based on the stochastic arithmetic to find the optimal number of iterations in...
متن کاملDistributed Fuzzy Adaptive Sliding Mode Formation for Nonlinear Multi-quadrotor Systems
This paper suggests a decentralized adaptive sliding mode formation procedure for affine nonlinear multi-quadrotor under a fixed directed topology wherever the followers are conquered by dynamical uncertainties. Compared with the previous studies which primarily concentrated on linear single-input single-output (SISO) agents or nonlinear agents with constant control gain, the proposed method is...
متن کاملCooperation without Exploitation between Self-interested Agents
We study how two self-interested agents that play a sequence of randomly generated normal form games, each game played once, can achieve cooperation without being exploited. The agent learns if the opponent is willing to cooperate by tracking the attitude of its opponent, which tells how much the opponent values its own payoff relative to the agent’s payoff. We present experimental results obta...
متن کامل