Coupled Sensor Configuration and Path-Planning in Uncertain Environments using Multimodal Sensors

St. Laurent, Chase

Etd

Coupled Sensor Configuration and Path-Planning in Uncertain Environments using Multimodal Sensors

Public

This dissertation develops a method for coupling sensor configuration and path-planning, two tasks often decoupled, as a means to reduce an agent's exposure to unknown but observable threats. This method exploits task-driven sensor configuration for gathering information with statistical relevance to the path-plan, the region of interest. The method addresses a sensor's field of view and the effect of overlapping field of views. Additionally, the sensor configuration optimizes the trade-off between sensor quality and quantity of information. We show that by coupling sensor configuration with path-planning in this manner, a near-optimal path-plan for the agent can be discovered with fewer observations than information-greedy approaches. Several scenario-specific variations are devised, such as greedy batch and exploration efficient modifications which handle cases where the path-plan optimality requirement is strict and when there are many sensors available, respectively. Extensions to multi-agent multi-goal scenarios are provided. We address computational performance through qualitative sensor configuration, which is achieved with cluster analysis, yielding suboptimal fast-approximations to sensor configuration for path-planning. These topics are decorated with numerical studies and examples displaying their benefits and performance guarantees. Building upon the prior topics, a self-adaptive surrogate optimization function was devised which enables sequential sensor configuration with near-optimality guarantees. The adaptive component balances the exploration-exploitation trade-off in discovering the optimal path-plan. We utilize the sequential sensor configuration for scenarios such as heterogeneous sensor configuration whereby the threat environment is multimodal. We consider scenarios in which there may be a penalty for waiting for an optimal-path plan and develop an active sensor configuration strategy which can be applied to dynamic and evolving spatiotemporal threat environments.

Creator