Smart Mobile Crowdsensing With Urban Vehicles: A Deep Reinforcement Learning Perspective

Abstract

Mobile crowdsensing (MCS) is a promising sensing paradigm based on the mobile node which provides the solution with cost-effectiveness to perform urban data collection. To monitor the urban environment and facilitate the municipal administration, more and more applications adopt vehicles as participants to carry out MCS tasks. The performance of the applications highly depends on the sensing data which is influenced by the recruiting strategy on vehicles. In this paper, we propose a novel vehicle selection algorithm to maximize the sensing range with limited cost while the vehicle selection problem was proved to be NP-complete. Specifically, we modeled the interaction between MCS server and candidate vehicles as a Markov decision process and formulated the maximum spatial temporal coverage (STC) optimization as a deep reinforcement learning process. The performance of our deep reinforcement learning-based vehicle selection (i.e., DRLVS) algorithm is evaluated with real trajectory dataset. The numerical result indicates that the proposed algorithm achieves an optimal solution and maximizes the STC.

Matching Learning-Based Relay Selection for Substation Power Internet of Things

“…Step 2: Iterative Matching (9) while ∃ϕ(s i ) � ∅ do (10) for s i ∈ S do (11) ∀ s i ∈ S makes a request to its most preferred r j according to F i . (12) end for (13) Add the relay nodes selected by more than one source node into Ω (14) if Ω ≠ ∅ then (15) for r j ∈ Ω do (16) r j raises its price as (17).…”

“…To fully utilize the spectrum and energy resources, relay selection needs to be optimized dynamically according to the network state and service requirements. However, relay selection optimization in SPIoT still faces several critical challenges as below [10].…”

A Vehicular Crowdsensing Market for AVs

“…However, in our case we found the epochs Nash Equilibrium for all the vehicles working for a given task. Following a reinforcement learning approach Wang et al [16] model the problem of vehicle recruitment for spatio-temporal coverage purposes as a Markov Decision Process. In the reinforcement learning approach we have an agent who observes the system, it takes actions to maximize the long term reward, and these actions take the agent to a new stage.…”

Path Planning Method With Improved Artificial Potential Field—A Reinforcement Learning Perspective

“…Wang et al formulate the maximum spatial-temporal coverage optimization issue as a deep reinforcement learning process. A deep reinforcement learning based vehicle scheduling is adopted to produce an optimal solution and maximize the spatial-temporal coverage [33]. Wei et al train a deterministic policy gradient algorithm on an abstracted structure to imitate the deformation of the path under the external force.…”