Collision-free indoor navigation using an artificial potential field controller

Abstract

Path planning is the problem of finding a collision-free path from a mobile agent's initial position to its goal position, and moving the agent along that path. This Master's thesis presents a control algorithm based on artificial potential fields for navigating a mobile agent in indoor environments. An artificial potential field is a scalar field defined in the environment of the agent, whose intensity relates to objects in the environment and the goal of the agent. The gradient of the scalar field is a vector field and determines the motion of the agent. The proposed algorithm produces control vectors that move the agent in an indoor environment towards its goal position. Collision with obstacles is quantified probabilistically, and the resulting quantity called the probability of collision is used to adapt the potential fields. Localization is a necessary step for calculating the probability of collision and the control vectors. Anchor nodes with known positions in the environment provide measurement information necessary for localization. The agent performs range measurements relative to the anchor nodes, and its position is estimated by a cubature Kalman filter from observed data. The probability of collision is used to adapt the artificial potential field such that within the framework of our approximation of the true probability of collision, the resulting control vector attempts to keep the probability of collision under a threshold. Simulations show that the proposed algorithm enables the agent to navigate indoor environments without collisions in various scenarios. In the absence of local minima between the initial and goal positions of the agent, a path is easily bound with little to no adaptation of the fields. The algorithm also shows some capability for escaping from and avoiding local minima.