Actively Coupled Sensor Configuration and Planning in Unknown Dynamic Environments
Prakash Poudel, Jeffrey DesRoches, Raghvendra V. Cowlagi
TL;DR
Actively coupled sensor configuration and planning (A-CSCP) addresses path planning for an ego vehicle in an unknown, time-varying threat field using a mobile sensor network. The approach couples sensor placement to planning via context-relevant mutual information (CRMI) and adds a sensor-reconfiguration cost, with a term encouraging sensor proximity to the ego, to guide a greedy, iterative update of sensor configurations while the ego moves and replans. The threat field is modeled as a Gaussian-basis expansion $c(\boldsymbol{x},t)=1+\boldsymbol{\Phi}^\top(\boldsymbol{x})\boldsymbol{\Theta}(t)$ with linear-Gaussian dynamics for $\boldsymbol{\Theta}$ and measurements $\boldsymbol{z}=H(\boldsymbol{q})\boldsymbol{\Theta}+\boldsymbol{\eta}$, enabling a Kalman/UKF estimator and a closed-form CRMI expression. Numerical simulations show that maximizing CRMI while penalizing sensor travel and promoting ego proximity yields lower path exposure and more efficient sensor placement than alternative schemes, demonstrating real-time adaptability in unknown dynamic environments.
Abstract
We address the problem of path-planning for an autonomous mobile vehicle, called the ego vehicle, in an unknown andtime-varying environment. The objective is for the ego vehicle to minimize exposure to a spatiotemporally-varying unknown scalar field called the threat field. Noisy measurements of the threat field are provided by a network of mobile sensors. Weaddress the problem of optimally configuring (placing) these sensors in the environment. To this end, we propose sensor reconfiguration by maximizing a reward function composed of three different elements. First, the reward includes an informa tion measure that we call context-relevant mutual information (CRMI). Unlike typical sensor placement techniques that maxi mize mutual information of the measurements and environment state, CRMI directly quantifies uncertainty reduction in the ego path cost while it moves in the environment. Therefore, the CRMI introduces active coupling between the ego vehicle and the sensor network. Second, the reward includes a penalty on the distances traveled by the sensors. Third, the reward includes a measure of proximity of the sensors to the ego vehicle. Although we do not consider communication issues in this paper, such proximity is of relevance for future work that addresses communications between the sensors and the ego vehicle. We illustrate and analyze the proposed technique via numerical simulations.