Online Robot Trajectory Optimization for Persistent Environmental Monitoring

Abstract

This paper presents a control-theoretic approach for trajectory optimization of mobile robots suitable for environmental monitoring. The method is based on the optimization of the Constructability Gramian in order to maximize the information collected while traversing a state trajectory. The maximization of the collected information is combined with energy constraints to define an optimization-based controller that achieves persistent environmental monitoring. The results of its application to the estimation of the concentration of a diffusing gas using a mobile robot are shown in simulation.

Multi-Robot Persistent Environmental Monitoring Based on Constraint-Driven Execution of Learned Robot Tasks

“…(i) We extend the persistent environmental monitoring approach developed in [9] to the multi-robot case by learning a coordinated-control policy using the RL paradigm which leverages the control-theoretic measures of information collected in the environment (ii) We show how to execute the learned multi-robot policy by means of the RL paradigm in a constraint-driven control framework, which allows us to combine the environmental monitoring task with energy control in order to achieve persistent environmental monitoring…”

“…The approach presented in this paper is based on the one proposed in [9], where a persistent environmental monitoring strategy is proposed, which consists in optimizing the robot trajectory in order to maximize the information gathered while traversing it. This approach has two advantages: (i) It possesses theoretical guarantees in terms of information collected to reconstruct an environmental field of interest, and (ii) the environment monitoring task lends itself to be embedded in a constraint-driven control framework.…”

“…The environmental monitoring approach in [9] based on active sensing, however, assumes the robotic system to be differentially flat in order to compute the control input required to track the optimized state trajectory. In general, this is not the case when considering multi-robot systems controlled in a coordinated fashion [10].…”

“…In general, this is not the case when considering multi-robot systems controlled in a coordinated fashion [10]. Moreover, compared to the single-robot case considered in [9], where the measurements collected by a robot are used by itself to plan the most informative trajectory, in multi-robot scenarios it is not clear how collected measurements should be utilized by the team of robots to estimate an environmental field of interest in a distributed fashion.…”

“…In this paper, leveraging the insights of the controltheoretic active sensing approach, we propose to extend the approach proposed in [9] by considering a multi-robot setting. This is achieved by learning a coordinated-control policy for multi-robot environmental monitoring tasks using the reinforcement learning (RL) paradigm.…”