PaperlayerView exact revision on arXiv

Source-linked AI summary

Resource Allocation and Service Provisioning in Multi-Agent Cloud Robotics: A Comprehensive Survey

Mahbuba Afrin, Jiong Jin, Akhlaqur Rahman, Ashfaqur Rahman, Jiafu Wan, Ekram Hossain

arXiv:2104.14270v1cs.ROcs.DC

TL;DR

Robots have limited computational and storage capacity, and coordinating resource allocation across heterogeneous resources is challenging. This paper surveys resource allocation and service provisioning in multi-agent cloud robotics, organizing research on pooling, offloading, and scheduling. It presents a taxonomy, lessons learned, a holistic framework, and research directions.

  • Problem

    Robots are limited in computation, storage, and energy, while allocating tasks across heterogeneous resources in multi-agent cloud robotics is challenging.

    Source: §I · p. 1, A. Multi-Agent Cloud Robotics and… · p. 2

  • Method

    The paper surveys resource allocation and service provisioning, discussing resource pooling, computation offloading, and task scheduling and presenting a taxonomy and holistic framework.

    Source: C. Motivation of this Article and Our… · p. 4, §VII · p. 21

  • Results

    The survey reviews existing policies and summarizes lessons learned in a holistic framework with potential research directions.

    Source: §VII · p. 21

  • Takeaways & Limitations

    The survey aims to serve as a reference and provide guidelines for resource allocation and service provisioning in multi-agent cloud robotics.

    Source: §VII · p. 21

Abstract

from arXiv · show

Robotic applications nowadays are widely adopted to enhance operational automation and performance of real-world Cyber-Physical Systems (CPSs) including Industry 4.0, agriculture, healthcare, and disaster management. These applications are composed of latency-sensitive, data-heavy, and compute-intensive tasks. The robots, however, are constrained in the computational power and storage capacity. The concept of multi-agent cloud robotics enables robot-to-robot cooperation and creates a complementary environment for the robots in executing large-scale applications with the capability to utilize the edge and cloud resources. However, in such a collaborative environment, the optimal resource allocation for robotic tasks is challenging to achieve. Heterogeneous energy consumption rates and application of execution costs associated with the robots and computing instances make it even more complex. In addition, the data transmission delay between local robots, edge nodes, and cloud data centres adversely affects the real-time interactions and impedes service performance guarantee. Taking all these issues into account, this paper comprehensively surveys the state-of-the-art on resource allocation and service provisioning in multi-agent cloud robotics. The paper presents the application domains of multi-agent cloud robotics through explicit comparison with the contemporary computing paradigms and identifies the specific research challenges. A complete taxonomy on resource allocation is presented for the first time, together with the discussion of resource pooling, computation offloading, and task scheduling for efficient service provisioning. Furthermore, we highlight the research gaps from the learned lessons, and present future directions deemed beneficial to further advance this emerging field.

I. INTRODUCTION

Multi-agent cloud robotics combines robots, edge nodes, and cloud data centres to support robotic applications that standalone robots cannot compute at large scale. Its heterogeneous resources and diverse task requirements make coordinated resource allocation and service provisioning challenging.

I. INTRODUCTION

  • Robots have limited capacity for large-scale computation, while multi-robot collaboration faces heterogeneous energy capacity, mobile communication, and limited storage.Cloud robotics extends multi-robot systems by using virtual machines or containers and resources from local robots and remote data centres.Source: §I · p. 1
  • Multi-agent cloud robotics integrates robot, edge, and cloud computing to address long communication delays between robots and distant cloud data centres.The passage places soft real-time tasks such as image processing and path planning at the edge, while latency-tolerant, storage-heavy tasks go to the cloud.Source: §I · p. 1
  • Distributed resources can reduce communication delay for CPS data, lessen reliance on distant cloud data centres, and support diverse applications and task classification.The introduction also identifies reduced bandwidth requirements as a benefit of bringing resources closer to robots.Source: §I · p. 2

A. Multi-Agent Cloud Robotics and Related Computing Paradigms

  • The paradigm differs from MCC and MEC through its n-tier use of local, proximal, and global computing agents, support for autonomous robot decisions, and dynamic robot mobility.MCC uses cloudlets between phones and data centres, while MEC provides virtual servers at cellular base stations.Source: A. Multi-Agent Cloud Robotics and… · p. 2
  • Resource allocation assigns applications to capable resources, while service provisioning establishes resource pooling and offloading and assists task scheduling.Their policies must coordinate heterogeneous resources and accommodate diverse requirements in this synthesis of robotics, tele-operation, edge computing, and cloud technologies.Source: A. Multi-Agent Cloud Robotics and… · p. 2

B. Challenges of Resource Allocation and Service Provisioning in Multi-Agent Cloud Robotics

Resource allocation and service provisioning face system-specific QoS, energy, security, coordination, and infrastructure challenges in multi-agent cloud robotics. The survey organizes prior work and identifies gaps and future directions.

B. Challenges of Resource Allocation and Service Provisioning in Multi-Agent Cloud Robotics

  • Edge constraints on installing resource-enriched devices limit compute-intensive AI and machine learning, requiring compatible-resource classification and selection.The paper also identifies the difficulty of developing a comprehensive business model for robotic systems’ varied service requirements.Source: B. Challenges of Resource Allocation… · pp. 3–4
  • Resource policies must place complex sensor-stream processing in real time and support coordinated cooperation among heterogeneous robots, including under limited communication.Distributed or federated learning adds resource-management overhead to collaboration.Source: B. Challenges of Resource Allocation… · p. 3
  • Policies must address cross-provider interoperability and synchronize decision-making across robots, edge, and cloud rather than rely on a single decision maker.Resources may use separate black-box systems for sharing, privacy, and fault tolerance.Source: B. Challenges of Resource Allocation… · p. 3
  • Robots’ changing mobility and heterogeneous processing, networking, and energy characteristics complicate the on-demand computation–communication trade-off in offloading.Offloading can give low-energy or low-power robots access to edge and cloud resources, but requires additional communication time.Source: B. Challenges of Resource Allocation… · p. 3
  • System-specific QoS and residual-energy constraints require tailored, time-efficient resource selection; edge resources also face availability, energy, and deployment-cost limits.The service deadline may be stricter in a smart hospital than on a smart farm, while accuracy or cost priorities differ across applications.Source: B. Challenges of Resource Allocation… · pp. 3–4
  • Shared robot, edge, and cloud resources increase exposure to security threats, while infrastructure failures and interference make fault tolerance difficult.The paper notes that MCC and MEC techniques cannot be directly transferred to this setting.Source: B. Challenges of Resource Allocation… · p. 4

C. Motivation of this Article and Our Contribution

  • The survey is presented as the first review jointly covering resource allocation and service provisioning in multi-agent cloud robotics.It aims to help readers understand the state of the art and use cases across application domains.Source: C. Motivation of this Article and Our… · p. 4
  • The survey presents a resource-allocation taxonomy and examines resource pooling, computation offloading, and task scheduling for service provisioning.It also identifies research gaps and offers future directions with a holistic framework.Source: C. Motivation of this Article and Our… · p. 4

D. Article Organization

The article reviews current practices and applications of multi-agent cloud robotics, compares its survey with related work, and proceeds from an overview to resource-allocation topics and future directions.

D. Article Organization

  • The article moves from an overview and comparison with related surveys to resource allocation, pooling, offloading, scheduling, research gaps, and a prospective framework.It concludes with future research directions and concluding remarks.Source: D. Article Organization · pp. 4–5

II. OVERVIEW OF MULTI-AGENT CLOUD ROBOTICS ADVANCEMENTS

  • The overview covers applications and recent practices in academia and industry, alongside a comparison with existing surveys..Source: §II · p. 5

A. Current and Future Applications of Multi-Agent Cloud Robotics

  • Multi-agent cloud robotics is used across smart factories, remote healthcare, smart farms, and disaster management for tasks including SLAM, vision, navigation, and surgical assistance..Source: A. Current and Future Applications of… · p. 5

1) Industry 4.0:

  • Industry 4.0 applications include industrial navigation, product management, customer maintenance, and material handling, but heterogeneous tasks and changing workloads make task and computational-load scheduling key challenges.The paper identifies efficient resource-allocation strategies as a fundamental demand for these applications.Source: 1) Industry 4.0 · p. 5

2) Health Care:

The surveyed application areas include healthcare, agriculture, and disaster management, alongside industrial efforts to integrate multi-robot systems with edge and cloud computing.

2) Health Care:

  • Healthcare robotics supports remote monitoring, elderly care, mobility and communication assistance, and personalized services; time-critical tasks require resource allocation for QoS.Examples include indoor localization, care reminders, environmental monitoring, and domiciliary medical reminders.Source: 2) Health Care · p. 5

3) Agriculture:

  • Smart farming combines sensors, IoT, GPS, and aerial and ground robots for monitoring and agricultural services, while unpredictable conditions and events remain a challenge.Robotic and sensor data must support agro-services despite communication constraints; the paper describes these technologies as still in their infancy in agriculture.Source: 3) Agriculture · p. 5

4) Disaster Management:

  • Disaster response uses robot, edge, and cloud resources toward a central mission, requiring task allocation and offloading that account for environmental uncertainty.The cited applications include human-rescuer collaboration, UAV emergency monitoring, large-scale search and rescue, and fire-driven emergency management.Source: 4) Disaster Management · pp. 5–6

B. Recent Industrial Practices of Multi-Agent Cloud Robotics

  • Technology and business organizations are developing platforms and services that integrate multi-robot systems, edge computing, and cloud computing.Examples in the passage include Google Cloud Robotics, AWS RoboMaker, ROS, Honda RaaS, and CloudMinds.Source: B. Recent Industrial Practices of… · p. 6
  • Efficient resource allocation and service provisioning remain concerns despite industrial initiatives, and the survey systematically summarizes related research..Source: B. Recent Industrial Practices of… · p. 6

C. Related Surveys

The section distinguishes this survey from prior reviews and introduces resource allocation and computational resources as foundations for the taxonomy.

C. Related Surveys

  • Prior surveys cover robot assignment, cloud provisioning, mobile offloading, or cloud-robotics architecture, but none jointly surveys pooling, offloading, and scheduling for multi-agent cloud robotics.The paper notes that resource allocation challenges in multi-agent cloud robotics were not addressed by the cited surveys.Source: C. Related Surveys · p. 6

III. RESOURCE ALLOCATION IN MULTI-AGENT CLOUD ROBOTIC SYSTEMS

  • Resource allocation assigns a resource to a task based on availability and QoS, while the relevant parameters and resource availability vary across applications and over time.The survey’s taxonomy covers resource type, performance parameter, application structure, service model, allocation mechanism, and evaluation method.Source: §III · pp. 6–7

A. Resource Type

1) Computational resource:

  • Cloud instances support large-scale computation, but public, private, and hybrid cloud arrangements have different access, security, cost, and task-placement trade-offs.Hybrid cloud can assign mission-critical or real-time tasks to private instances and delay-tolerant tasks to public instances, though data dependencies complicate management.Source: 1) Computational resource · p. 7
  • Robots offer comparatively low transfer delay but have constrained onboard computing; edge and fog resources process tasks near robots, though their availability, energy, and deployment costs constrain selection.Cloud resources offer greater computing power but add data-transfer delay, making resource choice challenging for latency-sensitive tasks.Source: 1) Computational resource · pp. 7–8
  • The taxonomy figure organizes resource allocation in multi-agent cloud robotics..Source: Figure 2 · p. 8

2) Network resource:

Network resources support connectivity and data flow, while storage can be placed in cloud, edge, or robot memory. Allocation must account for differing network and storage needs, including latency, battery, privacy, and access constraints.

2) Network resource:

  • Network resources include bandwidth, networking nodes, network slices, and wireless power, shared through mesh orchestration to maintain data flow.The passages identify bandwidth, slices, and power transmitters as allocated network resources.Source: 2) Network resource · p. 8
  • Bandwidth allocation supports multisensor-data retrieval and sharing, network communication, and QoS for services from cloud and edge instances.The cited works – focus on bandwidth allocation in multi-agent cloud robotics.Source: 2) Network resource · p. 8
  • Networking nodes include routers, switches, antennas, access points, and base stations; their placement matters for timely data exchange.Robots themselves are also treated as networking nodes in.Source: 2) Network resource · p. 8
  • Wireless power can supplement robots’ short-lived batteries, but mutual interference is a central transmission challenge.Remote robots otherwise may need to return to a base station, replace batteries, or use tethering; transfer methods include radiative and non-radiative techniques –.Source: 2) Network resource · p. 8
  • Network slicing can serve robots with heterogeneous requirements, such as highly reliable transfer or low latency with high data rate.Slice composition, resource allocation, and distribution across robotic systems remain open questions, and wireless power allocation and slicing mechanisms are described as early-stage research.Source: 2) Network resource · p. 8

3) Storage:

  • Storage is available through cloud servers, foglets, and robot memory, but most existing work emphasizes computation and network allocation over storage.Selecting data and its storage location is important for ease of access during real-time interactions.Source: 3) Storage · pp. 8–9
  • Cloud storage offers extensive capacity and collective-learning support, but real-time exchange requires uninterrupted internet and sensitive applications require privacy-aware infrastructure selection.Cloud storage is maintained by providers and can use virtualization and NoSQL databases.Source: 3) Storage · p. 9
  • Robot memory supports local participation beyond communication range, but limited capacity and reconfiguration make it better suited to static settings than dynamic ones.Edge storage can reduce latency for hard real-time applications, while its security and privacy require attention.Source: 3) Storage · p. 9

B. Performance Metrics

Resource-allocation performance is assessed through time, power, cost, and QoS-related measures. The section describes timing and energy demands alongside cost trade-offs for providers and users.

1) Time:

  • Time performance concerns minimizing computation and communication time and completing tasks within their deadlines.Computation time depends mainly on assigned-resource speed and task size; communication time is networking delay among system entities.Source: 1) Time · p. 9
  • Deadlines distinguish latency-sensitive from latency-tolerant applications and guide resource allocation toward application QoS requirements.Resource discovery time, service access time, and sensor data-sensing frequency are additional time indicators.Source: 1) Time · p. 9

2) Power:

  • Energy consumption depends on task computation time and resource energy use, while heat rises with resource use and increases hardware-failure probability and carbon footprint.Battery lifetime and communication-medium energy characteristics also merit consideration, but these areas remain subject to further research.Source: 2) Power · p. 9

3) Cost:

  • Cost-aware allocation balances provider profit and user QoE within budgets through resource deployment costs, user expenses, and service pricing.Deployment cost covers infrastructure, sensors, robots, VMs, containers, and networking; users pay for resource use, while provider profit depends on charges and business costs.Source: 3) Cost · pp. 9–10

4) Quality-of-Service:

QoS reflects application requirements under service agreements and depends on resource availability and fault tolerance. Robotic tasks can be modeled as dependent workflows or independent task bags, while service provisioning also requires suitable infrastructure.

4) Quality-of-Service:

  • QoS distributes resources according to application requirements under the provider–robotic-system SLA, with availability and fault tolerance also shaping service quality.Anomalies including link failures, power shortages, limited bandwidth, and interference can disrupt simultaneous execution.Source: 4) Quality-of-Service · p. 10

C. Application Structure

2) Bag-of-Tasks:

  • Resource allocation must account for multi-tenancy and application delay sensitivity, even when that requires crossing architectural differences.Source: 2) Bag-of-Tasks · p. 10

D. Service Model

1) IaaS:

  • IaaS pools virtualized resources such as VMs, containers, servers, and storage, but must accommodate heterogeneous robots and evolving requirements.The infrastructure should support flexible resource distribution and extension with functions such as self-healing and fault tolerance.Source: 1) IaaS · pp. 10–11

2) PaaS:

The survey describes platform, software, robot, and networking services, alongside resource-allocation mechanisms. It notes that optimization choices depend on system objectives and constraints, and may be infeasible in real time for complex problems.

2) PaaS:

  • PaaS provides operating systems, runtimes, interfaces, databases, and web servers, including platforms for map optimization, robot patrol, and inference or learning.Edge-enabled PaaS feasibility is investigated in,,, while offers inference and learning for grasp planning and object recognition.Source: 2) PaaS · p. 11

3) SaaS:

  • SaaS delivers on-demand robotic applications such as object recognition, path planning, map building, and speech translation over the internet.Examples include cloud-migrated software packages for multiple robots and cloud-parallelized robotic algorithms [135].Source: 3) SaaS · p. 11

4) RaaS:

  • RaaS dynamically combines distributed robots as services for computing, sensing, motion planning, navigation, and perception.Robots and sensors can expose functions to a virtualized pool, while onboard sensors provide sensing-as-a-Service –.Source: 4) RaaS · pp. 11–12

5) NaaS:

  • NaaS uses nearby drones or UAVs as repeaters, range extenders, base stations, access points, data ferries, or aerial cloudlets.Efficient provisioning must address service-robot energy use and UAV flying time through trajectory planning.Source: 5) NaaS · p. 12

E. Mechanism

  • Resource-allocation mechanisms include formal optimization, approximation, evolutionary computing, game theory, heuristics, and deep learning.The mechanism selected depends on the system environment and application requirements.Source: E. Mechanism · p. 12

1) Formal optimization:

  • Optimization methods are chosen according to objective-function and constraint characteristics, but complex problems may lack feasible solutions under real-time requirements.For example, linearizes a mixed-integer nonlinear problem into an ILP model and reports reduced computation and communication time and minimal latency.Source: 1) Formal optimization · p. 12

2) Approximation method:

The section surveys resource-allocation approaches and their trade-offs, then frames resource pooling and coordinated service provisioning as foundations for managing robotic tasks across local, edge, and cloud resources.

2) Approximation method:

  • Heuristics can find near-optimal allocations faster, but do not guarantee the best optimum; reported examples target computation, storage, and bandwidth.Reported examples include a 77% time-and-cost reduction using local resources only, 99.3% time savings with 23.8% lower energy use, and a 22% better near-optimal solution.Source: 2) Approximation method · p. 12

3) Evolutionary computing:

  • Genetic algorithms optimize multiple objectives but may miss the most optimal solution, while NSGA-II addresses multi-objective resource allocation; population-based search can be too slow for real-time use.Reported results include up to 10% lower computation time and cost, QoS satisfaction for 98% of tasks, and up to 18% improvement in optimizing time, energy, and cost.Source: 3) Evolutionary computing · pp. 12–13

4) Game theory:

  • Game-theoretic allocation addresses competition for shared resources, but depends on rational decisions and can be difficult to model or solve in real time under communication constraints.A testbed-and-simulation study reports auction-based bandwidth allocation increasing bandwidth usage by 35.9%.Source: 4) Game theory · p. 13

5) Deep learning:

  • Deep reinforcement learning is proposed for discrete and continuous resource-allocation states and actions, but its feasibility depends on clearly defining the reward or objective and system constraints.In one simulation, reinforcement learning performs better than a genetic algorithm when cloud computing resources are limited.Source: 5) Deep learning · p. 13
  • Reliable, delay-aware, and energy-efficient allocation remains important for heterogeneous robots, while most policies are simulation-evaluated and few use real-world testbeds.The survey also identifies faster processing as an inherent need of real-world robotic applications.Source: 5) Deep learning · p. 13

IV. SERVICE PROVISIONING IN MULTI-AGENT CLOUD ROBOTIC SYSTEMS

  • Service provisioning comprises resource pooling, robot-selected task offloading, and resource-manager scheduling by QoS requirements such as deadlines before resources are allocated.Robots send application profiles and receive resource specifications; the manager provisions resources and schedules received tasks according to availability and QoS.Source: §IV · pp. 13–14

A. Resource Pooling

  • Resource pooling arranges resources for processing and storage, but resource selection must address security, privacy, trust, communication efficiency, and resilience to wireless failures.The survey notes physical-resource protection, confidentiality, authentication, and privacy alongside protocol choice, route discovery, and backup mechanisms.Source: A. Resource Pooling · pp. 14–15

1) Resource Discovery:

Resource discovery matches resource arrangements to workload patterns, while resource scaling adapts dispersed resources; computation offloading must also account for robot mobility and communication costs.

1) Resource Discovery:

  • On-demand discovery suits frequently changing workloads but costs more, whereas reservations reduce expenses for predictable workloads but risk under- or over-provisioning; spot-instance discovery remains unexplored.Spot-instance prices vary with supply and demand, and systems bid for access.Source: 1) Resource Discovery · p. 15

2) Resource Scaling:

  • Resource scaling can aggregate like resources, amalgamate different capabilities, or shift resources between devices, with compatibility and communication time constraining what can be moved or combined.Aggregation is limited to augmentation of the same resource type; amalgamation adds heterogeneity and cross-platform management overhead.Source: 2) Resource Scaling · p. 15

B. Computation Offloading

  • Robot mobility distinguishes multi-agent cloud-robotic offloading from mobile-cloud offloading, and local robots can also provide resources; decisions must balance delay, mobility, data-transfer, and computation costs.The survey says mobile-cloud techniques cannot be directly applied to multi-agent cloud robotics.Source: B. Computation Offloading · pp. 15–16

1) How offloading decision should be made?:

Offloading decisions may be reactive or predictive, and their implementation trades off communication, computation, mobility, and coordination across robots and remote resources.

1) How offloading decision should be made?:

  • Reactive offloading decides after a need arises and requires speed, while proactive offloading predicts demand but adds decision time and computation overhead.The survey calls for balancing the two decision styles.Source: 1) How offloading decision should be… · p. 16

2) How actual offloading should be conducted?:

  • Partial offloading distributes application parts for parallel processing but adds partitioning, dependency, and aggregation costs; full offloading simplifies decisions while increasing bandwidth and single-resource energy use.The survey characterizes the choice as a trade-off between computation and communication, with energy-delay optimization also needed.Source: 2) How actual offloading should be… · p. 16
  • Single-robot decisions require less communication but may lack a system-wide view and create single-robot failure risk; multiple robots can improve network lifetime and energy optimization but complicate synchronization and security.Robot selection is therefore essential to offloading decisions.Source: 2) How actual offloading should be… · p. 16
  • Most surveyed approaches fully offload applications to cloud, while few use partial offloading; studies vary in algorithms, objectives, and the number of participating robots.Reported objectives include energy, latency, movement cost, and bandwidth or communication cost.Source: 2) How actual offloading should be… · p. 17
  • Offloading can become inefficient when edge or cloud capacity is limited or robots converge on the same server; mobility, bandwidth, and local-versus-remote costs remain open decision challenges.The survey calls for further research on selecting approaches for arbitrary system contexts.Source: 2) How actual offloading should be… · pp. 16–17

C. Task Scheduling

Task scheduling may be static, with costs estimated before assignment, or dynamic, without prior time or cost estimation. The review highlights scheduling and load balancing challenges alongside broader lessons on task placement, energy, mobility, resource management, security, and QoS.

1) Static scheduling:

  • Static scheduling assigns tasks after set intervals, estimating each task’s communication and computation costs beforehand.The review describes this as one scheduling strategy in the literature.Source: 1) Static scheduling · p. 17

2) Dynamic scheduling:

  • Dynamic scheduling assigns tasks without prior time or cost estimates; scheduling policies remain understudied and difficult to implement without holistic resource monitoring.The review also notes limited evaluation of existing frameworks and challenges adapting them to decentralized architectures and real-world conditions.Source: 2) Dynamic scheduling · p. 18, §V · p. 18

V. LESSONS LEARNED

  • Latency-sensitive tasks fit robots or edge nodes, while compute- and storage-intensive tasks fit the cloud; partitioning applications accordingly adds aggregation overhead.Task suitability varies across heterogeneous infrastructure, so partitioning can match tasks to resources while requiring results to be aggregated at the receiver.Source: §V · p. 18
  • Remote robots and edge resources require substantial energy, while uncertain renewable supply complicates proactive resource allocation and service provisioning.Grid-based energy can be costly in remote locations, and reactive energy policies may not suit renewable-dependent environments.Source: §V · p. 18
  • Uneven robot mobility complicates centralized monitoring and mobility-driven provisioning; dynamically switching between centralized and robot-initiated approaches is preferred.Robot-initiated requests can add communication and data-management overhead, while centralized movement of computation requires network understanding.Source: §V · p. 18
  • Coexisting resource managers create synchronization problems; negotiation can help, but limits independent decisions, motivating fair and adjustable responsibility sharing.The review recommends distributing resource-management responsibilities across the robotic, edge, and cloud components.Source: §V · p. 18
  • Exposure across edge, core, and cloud communication layers raises security risks, while heavyweight security can slow real-time interactions.Lightweight features are preferred, though their ability to guarantee scalable solutions remains an open research issue.Source: §V · p. 18
  • Prioritizing QoS can reduce SLA violations and relinquish rates while improving service-provider economic benefits.The review notes that universally accredited business models remain undeveloped and that profit-driven QoS relaxation can erode user-provider trust.Source: §V · p. 18

VI. A HOLISTIC FRAMEWORK AND FUTURE RESEARCH DIRECTIONS

The proposed holistic framework uses a distributed Cross-infrastructure Resource Allocator and Service Provisioner (CRASP) to coordinate resource and service management across robotic, edge, and cloud levels. It places latency-sensitive tasks at the edge, compute-intensive tasks in the cloud, and identifies event-driven applications as a resource-allocation challenge.

VI. A HOLISTIC FRAMEWORK AND FUTURE RESEARCH DIRECTIONS

  • The framework integrates hardware and software components through distributed CRASP components, placing latency-sensitive tasks at the edge and compute-intensive tasks in the cloud.It presents the framework as a way to support end-to-end interaction and scalable integration of resource and service management policies.Source: §VI · pp. 18–19, Figure 4 · p. 19

A. Robotic System Level

  • At the robotic level, CRASP profiles applications, classifies tasks, decides whether to offload them, and dispatches tasks and data to edge or cloud resources.A dispatcher and negotiator perceives platform state and guides input transmission rates through the flow controller while remote processing proceeds.Source: A. Robotic System Level · p. 19

B. Edge Infrastructure Level

  • At the edge, CRASP discovers and allocates nearby nodes for latency-sensitive tasks on demand, forwarding scheduled tasks to the cloud when edge resources fail or run short.A network monitor updates status, and edge nodes may interact in ad-hoc clusters.Source: B. Edge Infrastructure Level · p. 19

C. Cloud Infrastructure Level

  • At the cloud level, CRASP schedules compute-intensive tasks and reserves resources through a resource pool, with a monitor tracking execution.The framework favors reservation over dynamic provisioning for these longer-running tasks.Source: C. Cloud Infrastructure Level · pp. 19–20

1) Event-driven resource allocation and service provisioning:

  • Event-driven applications may trigger other applications that need to run simultaneously, but resource constraints can disrupt this arrangement.The review calls for allocation and provisioning policies that address these event-driven requirements.Source: 1) Event-driven resource allocation… · p. 20

2) Energy-efficient resource consolidation and scaling:

The survey identifies research challenges for resource allocation and service provisioning in multi-agent cloud robotics, including energy management, offloading, network slicing, mobility, provider coordination, pricing, and security. It synthesizes resource pooling, computation offloading, and task scheduling, and presents a holistic framework and future research directions.

2) Energy-efficient resource consolidation and scaling:

  • Renewable-energy variability makes resource consolidation and scaling attractive, but changing network topology and added management overhead complicate their use.Policies must account for these effects when allocating resources and provisioning services.Source: 2) Energy-efficient resource… · p. 20

3) Balance between pro-active and re-active offloading decision:

  • Proactive offloading suits predictable static robot mobility, while reactive decisions address dynamics over shorter periods; predictive analytics can help balance both.The passage calls for a balance that responds to uncertain events with greater accuracy over a longer period.Source: 3) Balance between pro-active and… · p. 20

4) Allocation of network slices for 5G enabled multi-agent cloud robotics:

  • Network-slice allocation must adapt application priorities at runtime to changing needs, QoS requirements, and user expectations.The proposed direction extends to 5G and beyond for mission-critical applications using the Tactile Internet.Source: 4) Allocation of network slices for 5G… · p. 20

5) Mobility-as-a-Service:

  • Robot mobility, especially UAVs acting as portable access points or signal boosters, could address uneven connectivity, but mobility-as-a-service needs further research.Ad-hoc access-point orchestration must also account for interference from inconsiderate deployment.Source: 5) Mobility-as-a-Service · p. 20

6) Real-time resource augmentation:

  • Black-box interfaces between providers make provisioning administrative and time-consuming; interoperable policies could help while accounting for providers’ individual interests.Source: 6) Real-time resource augmentation · p. 20

7) Pricing model for resource consumption:

  • A pricing model is needed to support provider revenue, user incentives, and compensation for SLA violations, despite heterogeneous components and operating costs.Source: 7) Pricing model for resource… · p. 20

8) Lightweight security measures during data management:

  • Sensitive robotic data needs lightweight security that protects data management and supports real-time interaction, since existing measures can be computation-intensive.The passage cites blockchain and 128-bit asymmetric key cryptography as examples of measures that can slow real-time interactions.Source: 8) Lightweight security measures… · p. 20

VII. CONCLUSION

  • The survey reviews resource allocation and service provisioning, discusses pooling, offloading, and scheduling challenges, and presents a holistic framework with research directions.Source: §VII · p. 21
Loading 2104.14270v1…