Adaptive Multi-Agent Control with Dynamic Obstacle Avoidance in a Limited Region

This paper presents an adaptive multi-agent control strategy with dynamic obstacle avoidance. The strategy considers the limitation of the obstacle avoidance space which makes it more implementable in practical applications. Inspired by the coverage problem, the obstacle avoidance strategy is developed based on the Centroidal Voronoi Tessellation (CVT) technique with a special design of time-varying density function. The strategy is implemented by an adaptive controller, taking into account the output constraints and the time-varying system uncertainties. The controller is constructed through the function approximation technique based immersion and invariance (FATII) approach. The stability of the corresponding control system is established, and the validity of the proposed controller is tested by simulations.

I. Introduction

The research on multi-agent control problems has received a considerable attention in recent decades [1]–[4]. A detailed review of multi-agent control techniques can be found in [5]. The implementation of the aforementioned techniques on realistic multi-agent systems raises an obstacle avoidance issue in complex working environments. Regarding this issue, various control methods have been proposed including online optimal control technique [6], model predictive control methods [7], reinforcement learning [8], [9], and artificial potential field methods [10], [11]. In these methods however, the space for the obstacle avoidance is commonly unconstrained (see Fig. 1), which is impractical for several scenarios, i.e., the vehicle-intersection coordination, or multi-agent systems carrying objects cooperatively, which require agents to stay close to their neighbors. Moreover, limiting the avoidance space can contribute to maintain the communication efficiency by preventing the decay of communication signals among the agents’ network.