Planning Mobile Robot Behavior in an Uncertain Multi-agent Environment

Abstract

The paper proposes an approach for safe navigation when changing lanes in a road scenario. The aim of this approach is to create the baseline value of the decision making module for the lane changing task with multiple agents to avoid collisions. The paper describes three models of the project. Model-1 is a working method for implementing modeling. Model-2 is a way to build a decision-making module using safe zones. Model-3 is the proposed way to build a decision-making module using value functions.

The paper presents a multi-agent approach to adjust traffic lights depending on traffic situations to reduce average delay time. The traffic lights of each intersection are controlled by a mobile agent. This approach creates a classical non-stationary environment, as each agent's decision affects neighboring agents. Therefore, each agent must not only learn from past experience, but also take into account the decisions of its neighbors to rule out dynamic changes in the traffic network. Fuzzy Q-learning and game theory are used to develop new policies based on previous experiences and decisions of neighboring agents. The results obtained from the simulation show the advantage of the proposed method over fixed time management, fuzzy Q-learning and fuzzy Q-learning methods.

Keywords: intelligent agent, behavior planning, goal planning, action planning, agent simulation, simulation simulation, neural network, fuzzy Q learning