Dépôt numérique
RECHERCHER

Human-agent-robot teamwork coordination in FiWi based tactile internet infrastructures.

Téléchargements

Téléchargements par mois depuis la dernière année

Plus de statistiques...

Chowdhury, Mahfuzulhoq (2018). Human-agent-robot teamwork coordination in FiWi based tactile internet infrastructures. Thèse. Québec, Université du Québec, Institut national de la recherche scientifique, Doctorat en télécommunications, 232 p.

[thumbnail of Chowdhury, Mahfuzulhoq.pdf]
Prévisualisation
PDF
Télécharger (2MB) | Prévisualisation

Résumé

The demands of increasingly latency-sensitive applications create challenges for pervasive mobile devices/robots to execute the involved computation-intensive tasks in a resource-efficient manner. Cooperative human-agent-robot teamwork (HART) holds promise to serve as a powerful paradigm to tackle the challenges of real-time task execution of mobile devices/robots. Integrated fiber-wireless (FiWi) enhanced networks play a pivotal role in ensuring qualityof- service (QoS) for several HART-centric applications due to their coverage and capacity advantages. In this work, integrated FiWi access networks consist of optical fiber (Ethernet passive optical network or EPON) and wireless (wireless local area network or WLAN) Ethernet technologies, which are integrated with their cellular counterparts, namely, 4G Long Term Evolution Advanced (LTE-A), to give rise to FiWi enhanced LTE-A heterogeneous networks (HetNets). To unleash the full potential of HART task coordination over FiWi enhanced 4G networks, this thesis first provides a detailed study of recent progress, enabling technologies, and briefly describes important open research challenges. To render the human-to-robot task allocation process more efficient, this thesis presents a local and non-local human-to-robot task allocation scheme for FiWi-based infrastructures according to several key design parameters such as the availability, skill set, distance to task location, minimum task processing time, and remaining energy of robots. To reduce failures during task execution, a neighboring robot assisted failure reporting mechanism is also proposed. Our obtained results show that, compared with traditional priority-based schemes, a task execution time efficiency of 18% can be achieved in our proposed local and non-local human-to-robot task allocation scheme. Due to limited computing, energy, and storage resources, robots can not always meet the task execution time and energy consumption requirements of many delay-sensitive applications. To improve the energy efficiency of the selected host robot while satisfying a given task deadline, this thesis presents a collaborative task execution scheme, in which the sensing sub-task is conducted by a suitable host robot and the computation sub-task is offloaded onto one of the suitable collaborative nodes consisting of central cloud, cloudlets, and neighboring robots. The presented results demonstrate that for a typical task input size of 240 KB, the collaborative task execution scheme decreases the task response time by up to 8.75% and the energy consumption by up to 14.98% compared to the only host robot based non-collaborative task execution scheme. Taking the idea of task offloading a step further, task migration among mobile HART members has emerged as an important research topic to improve the quality of experience (QoE) of mobile users (MUs) by minimizing their task execution time. Task migration broadens the scope of conventional computation task offloading by not only transferring the task from an MU onto the cloud, but also from one cloud server to another one for execution. Note, however, that task migration incurs an additional migration delay. Hence, for a given task migration gain and latency overhead, the question of how and where an MU's task should migrate to is key. After describing the key features of physical vs. cognitive tasks and collaborative robot (cobot) vs. stand-alone robot types, this thesis next investigates the problem of whether and, if so, when and where a HART-centric task should be best migrated to. For resource-efficient task execution, a context-aware task migration scheme is presented, in which the suitable task migration decision is made by taking into account given task processing capabilities of cloud/cloudlet agents and cobots, task execution deadline, user mobility, energy consumption of involved collaborative robots (cobots) and mobile devices, and task migration latency. Our obtained results show that for a typical task input data size of 600 MB, the cobot-to-agent (c2a) task migration (cloudlet near task location) scheme exhibits up to 20% task response time and 23% energy efficiency improvements over the traditional task execution without migration scheme. The results also indicate that in the case of an agent node failure, intra-agent task migration offers a higher task response time gain than inter-agent migration. Furthermore, to improve QoS for executing multiple HART tasks, the development of real-time task scheduling mechanisms has emerged as an interesting research issue by taking different real-time HART task properties, failure avoidance, and task processing capabilities into account. Thus, to improve the HART task execution process, this thesis next presents a community- and latency-aware HART task assignment scheme by using real-time information about arriving task requests for both isolated and clustered robots/agents. More specifically, a suitable multi-task scheduling scheme is presented for task on- and offloading based HART task execution with task prefetching and fault tolerance capabilities. To reap the benefits from task prefetching for executing multiple HART tasks, this thesis develops a novel prefetchingaware bandwidth allocation scheme that copes with both conventional broadband and task offloading data traffic at the same time. Next, a comprehensive analytical model is presented to investigate the performance of our proposed community- and latency-aware task offloading scheme in terms of mean task service time, delay and power saving ratio, task prefetching time efficiency, task service time gain to overhead ratio, among others. Our obtained results show that for a typical system of 32 integrated optical network unit-mesh portal points (ONUMPPs) and a polling cycle time of 100 ms, our proposed task offloading scheme achieves up to 31.3% and 32.7% task completion time gain over the task onloading scheme for nearby and remote task execution, respectively. The results demonstrate that for a typical task offload input data size of 500 MB, our proposed community- and latency-aware task offloading scheme with task prefetching capability offers a 11% higher task service time gain to overhead ratio than a conventional fetching based scheme. Our findings also suggest that for failure avoidance, the proposed fault tolerance mechanism is more effective in the considered task offloading scheme than the alternative failure recovery mechanism. Given human users' different preferences for real-time HART task execution, e.g., lower delay and monetary cost, suitable HART task coordination has emerged as an important research problem, taking dynamically changing cloud agent/robot resources, network bandwidth utilization as well as delay-sensitive and delay-tolerant HART task properties into account. To cope with these challenges, this thesis explores the synergy between caching, computation, and communications for achieving cost-effective HART task execution. More precisely, to minimize task execution delay and monetary cost, this thesis presents a user preference-aware HART task coordination framework that selects the appropriate dedicated or non-dedicated robot and cloud agent for given caching and computing HART task execution requirements. To cope with varying bandwidth resources, this thesis proposes a proactive bandwidth allocation policy for the execution of both delay-sensitive and delay-tolerant HART tasks. To minimize the task execution delay of delay-sensitive users, our proposed delay cost saving (DCS) based scheme selects suitable actors by using both dedicated and non-dedicated actors. Conversely, to minimize the monetary cost for delay-tolerant policy users, our proposed monetary cost saving (MCS) scheme selects appropriate actors only from the set of dedicated actors. Furthermore, this thesis also presents a proactive bandwidth allocation scheme that assigns preemptive and non-preemptive bandwidth resources to DCS and MCS policy users, respectively. Unlike alternative approaches, our findings indicate that the maximum throughput and minimum delay (MTMD) based resource assignment policy is useful for both DCS and MCS policy users due to its minimum task execution time and monetary cost. Our obtained results show that for a typical number of 10 tasks and 8 available dedicated robots, the DCS (MTMD) policy exhibits a 30.5% higher task execution time saving ratio and a 63.6% lower monetary cost saving ratio than the MCS (MTMD) policy. Our proposed user preference aware HART task coordination policy thus represents a promising solution to reduce both task execution delay and monetary cost for emerging Tactile Internet applications.

Type de document: Thèse Thèse
Directeur de mémoire/thèse: Maier, Martin
Mots-clés libres: caching; cloud computing; computation offloading; collaborative computing; delay cost saving; dynamic bandwidth allocation (DBA); energy efficiency, failure avoidance; fiber-wireless (FIWI) enhanced networks; human-agent-robot teamwork (HART); human-to-robot communication (H2R); human-machine co-activity; internet-of-things (IoT); mobile-edge computing (MEC); monetary cost saving; tactile internet, task and resource scheduling; and task migration
Centre: Centre Énergie Matériaux Télécommunications
Date de dépôt: 09 avr. 2019 21:14
Dernière modification: 09 avr. 2019 21:14
URI: https://espace.inrs.ca/id/eprint/8006

Gestion Actions (Identification requise)

Modifier la notice Modifier la notice