Paper deep dive
DAG-Based QoS-Aware Dynamic Task Placement for Networked Multi-Stage Control Pipelines
Thien Tran, Jonathan Kua, Thuong Hoang, Minh Tran, Yuemin Ding, Jiong Jin
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 92%
Last extracted: 7/8/2026, 5:47:53 PM
Summary
This paper introduces a Directed Acyclic Graph (DAG)-based Quality-of-Service (QoS)-aware Dynamic Task Placement (DTP) framework for networked multi-stage control pipelines in industrial robotics. The framework formalizes sensing-perception-planning-control workflows as a DAG with hard-anchored and relocatable stages, optimizing a window-based cost function that balances end-to-end latency, deadline violation rates, hardware utilization, and placement switching penalties. By incorporating hysteresis and minimum dwell-time constraints, the DTP algorithm mitigates placement chatter and extends prior Adaptive Task Placement (ATP) controllers, enabling robust Control-Communication-Computing (3C) co-design within Cloud-Fog Automation architectures.
Entities (8)
Relation Signals (6)
Dynamic Task Placement (DTP) → formalizes → pipeline as DAG
confidence 96% · This pipeline is formalized as a DAG with task-level and node-level attributes for compute cost, communication delay, and feasible placement sets
Dynamic Task Placement (DTP) algorithm → uses → hysteresis
confidence 95% · a DTP algorithm with hysteresis and a minimum dwell-time bounds placement chatter.
Physical AI (PAI) → relieson → closed-loop visual-servoing pipelines
confidence 93% · Current Physical AI (PAI) relies heavily on closed-loop visual-servoing pipelines, whose perception and planning stages may become computationally intensive onboard due to complex models embedded on robots.
Dynamic Task Placement (DTP) → extends → Adaptive Task Placement (ATP)
confidence 91% · the formal multi-stage extension of the empirically validated edge-based QoS-aware ATP controller from previous work.
Cost function → combines → Deadline Violation Rate
confidence 90% · a window-based cost function combines tail end-to-end latency, deadline violation rate, hardware utilization, and a Hamming-distance switching penalty
Cloud-Fog Automation (CFA) → provides → unified vision
confidence 87% · The Cloud-Fog Automation (CFA) reference architecture [3, 2] articulates a unified vision for networked industrial collaboration across the cloud–fog–edge continuum
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Current Physical AI (PAI) relies heavily on closed-loop visual-servoing pipelines, whose perception and planning stages may become computationally intensive onboard due to complex models embedded on robots. In practice, offloading the perception task to on-site edges statically is inappropriate for latency-sensitive, precise industrial settings over a standardized industrial network. This emphasizes the importance of Control-Communication-Computing (3C) co-design in industrial automation: monolithic local execution saturates AI-accelerated machine and robot hardware, while static edge offloading exposes the control loop to network jitter. Existing adaptive task placement (ATP) controllers can partially address the gap by relocating a single pipeline stage on binary threshold rules, without a multi-stage model and an explicit cost on placement switching. In this Work-in-Progress (WiP) paper, we propose a directed acyclic graph (DAG) based quality-of-service (QoS)-aware dynamic task placement (DTP) framework for sensing-perception-planning-control pipelines in networked robotics. This pipeline is formalized as a DAG with task-level and node-level attributes for compute cost, communication delay, and feasible placement sets; over a small interpretable candidate set (fully local, static offload, hybrid), a window-based cost function combines tail end-to-end latency, deadline violation rate, hardware utilization, and a Hamming-distance switching penalty, and a DTP algorithm with hysteresis and a minimum dwell-time bounds placement chatter. Our WiP paper presents the theoretical framework, a structured qualitative analysis, and a two-phase simulation plus hardware-in-the-loop validation roadmap.
Tags
Links
- Source: https://arxiv.org/abs/2605.19887v1
- Canonical: https://arxiv.org/abs/2605.19887v1
Trouble viewing inline? Open PDF directly →
Full Text
21,465 characters extracted from source content.
Expand or collapse full text
DAG-Based QoS-Aware Dynamic Task Placement for Networked Multi-Stage Control Pipelines Thien Tran1, Jonathan Kua1, Thuong Hoang1, Minh Tran2, Yuemin Ding3, and Jiong Jin4 Abstract Current Physical AI (PAI) relies heavily on closed-loop visual-servoing pipelines, whose perception and planning stages may become computationally intensive onboard due to complex models embedded on robots. In practice, offloading the perception task to on-site edges statically is inappropriate for latency-sensitive, precise industrial settings over a standardized industrial network. This emphasizes the importance of Control–Communication–Computing (3C) co-design in industrial automation: monolithic local execution saturates AI-accelerated machine and robot hardware, while static edge offloading exposes the control loop to network jitter. Existing adaptive task placement (ATP) controllers can partially address the gap by relocating a single pipeline stage on binary threshold rules, without a multi-stage model and an explicit cost on placement switching. In this Work-in-Progress (WiP) paper, we propose a directed acyclic graph (DAG) based quality-of-service (QoS)-aware dynamic task placement (DTP) framework for sensing–perception–planning–control pipelines in networked robotics. This pipeline is formalized as a DAG with task-level and node-level attributes for compute cost, communication delay, and feasible placement sets; over a small interpretable candidate set (fully local, static offload, hybrid), a window-based cost function combines tail end-to-end latency, deadline violation rate, hardware utilization, and a Hamming-distance switching penalty, and a DTP algorithm with hysteresis and a minimum dwell-time bounds placement chatter. Our WiP paper presents the theoretical framework, a structured qualitative analysis, and a two-phase simulation-plus-hardware-in-the-loop validation roadmap. I Introduction The compute envelope of factory robotics is being reshaped by Physical AI (PAI) workloads. World Foundation Models (WFMs) for surrounding cognitive and agentic visual servoing demand inference costs that exceed the on-board envelope of typical industrial robots, while LLM-based task dispatchers exhibit cold-start latencies of sub-20 seconds, orders of magnitude beyond the sub-millisecond deadlines typical of industrial contact-rich control loops within Vision-Language-Action (VLA) models embedded on PAI [5, 4]. This mismatch demands a unified networked orchestration framework that binds heterogeneous AI capabilities to the robot–edge compute fabric within the closed-loop timing budget. The Cloud-Fog Automation (CFA) reference architecture [3, 2] articulates a unified vision for networked industrial collaboration across the cloud–fog–edge continuum, while deterministic substrates such as Time-Sensitive Networking (TSN) and 5G-URLLC [9] provide the underlying real-time fabric. Together, they make co-located edge servers a viable extension of the PAI compute envelope, provided placement decisions respect the closed-loop timing budget. However, the layer specifies the runtime decision rule: under workload and network variability that no design-time profile can predict, the deployment question shifts from “whether” to offload to “which” pipeline stages run, where by being decided online. A single static placement cannot serve this regime: both perception/planning compute demand and link delay are non-stationary, varying with scene complexity and model branching, and the cost of a wrong decision is asymmetric; a missed deadline propagates into actuation jitter that can violate safety envelopes. Empirical studies on industrial testbeds report deadline-violation rates above 40% under monolithic local deployment of visual-servoing workloads [7]. Adaptive placement is, therefore, a deployment necessity; the open question is what such a rule should observe what it should optimize, and how it should be stabilized for industrial deployment. DAG-based task offloading has been studied extensively in mobile edge computing (MEC) [8, 6], but these models target “batch workflows” and optimize average latency or energy rather than deadline violation rate (VDV_D) under closed-loop control. Our previous Adaptive task placement (ATP) on physical robot testbeds demonstrates the value of dynamic offloading, but each system relocates only a “single” pipeline stage using binary threshold rules on individual fixed metrics, providing no formal DAG model, no multi-objective cost function, and no analytical resistance to task placement oscillation. Figure 1: DAG-QoS DTP system architecture. (a) The four-stage sensing–perception–planning–control pipeline as a DAG: T1T_1, T4T_4 are hard-anchored; T2T_2, T3T_3 form the relocatable adaptation space. (b) Three candidate placements (LOC, SO, HYB) on the R1R_1/E/R2R_2 compute fabric. (c) Window-based DTP loop on E: per-window QoS observation, cost JkJ_k, and decision πk∗π^*_k with hysteresis Δmin _min and dwell-time NminN_min, maintaining VD≤5%V_D≤ 5\%. In this Work-in-Progress (WiP) paper, we propose a DAG-based QoS-aware DTP framework, the formal multi-objective extension of the empirically validated edge-based QoS-aware ATP controller from previous work. The system architectural framework is shown in Fig. 1. We present the complete theoretical framework, DAG model, cost function, and DTP algorithm, with the structured qualitative analysis and the validation roadmap. This work contributes the DAG pipeline model that distinguishes hard-anchored sensing and control stages from the relocatable perception and planning adaptation space, agnostic to the LAN/TSN/5G-URLLC substrate. On this model, we define the multi-objective QoS cost function and the window-based DTP algorithm with hysteresis and minimum dwell-time that bounds placement chatter, formalizing the 3C principles within the single per-window optimization. I Theoretical System and DAG-based Model I-A Compute Nodes and Workload DAG Let =ℛ∪ℰ∪N=R denote the global compute node set, comprising robots – ℛR, edge servers – ℰE, and (optionally or extensively) cloud nodes – C. The target empirical factory configuration is R1R_1 (light robot with monocular camera), R2R_2 (heavy robot driving a manipulator), and E (edge server), interconnected by an industrial real-time network with configurable delay and jitter. The framework is agnostic to the link technology, engineered LAN, TSN, or 5G-URLLC [9], and is parameterized by the per-edge delay distribution dij(⋅)d_ij(·). A control pipeline is modeled as a directed acyclic graph – G=(,ℰG)G=(V,E_G), with task set =T1,T2,T3,T4V=\T_1,T_2,T_3,T_4\ corresponding to four stages, and linear precedence ℰG=(T1,T2),(T2,T3),(T3,T4)E_G=\(T_1,T_2),(T_2,T_3),(T_3,T_4)\. Each task v∈v carries: • The feasible placement set v⊆N_v ; • The compute-time function cv:v→ℝ>0c_v:N_v _>0; and • The CPU-utilization contribution uv:v→[0,1]u_v:N_v→[0,1]. Denote that each precedence edge (vi,vj)∈ℰG(v_i,v_j) _G incurs a communication delay dij(ni,nj)≥0d_ij(n_i,n_j)≥ 0 depending on link characteristics and payload size of the associated operations. I-B Placement Mapping and Architectural Invariant A placement mapping π:→π:V assigns each task to a feasible node. The visual servoing pipeline imposes: T1 _T_1 =R1, =\R_1\, T4 _T_4 =R2, =\R_2\, (1) T2 _T_2 =R1,E, =\R_1,E\, T3 _T_3 =R2,E. =\R_2,E\. (2) This encodes the key architectural invariant of the multi-PAI System in agentic industrial automation operation: • Hard-anchored stages (T1T_1, T4T_4) remain on physical AI-accelerated machines and robots, preserving sub-millisecond local operations regardless of network state. • Relocatable adaptation space (T2T_2, T3T_3) migrates across the Cloud–Edge–Robotics continuum (native edge tiers) subject to the QoS cost function defined in Section I. I-C Control Loop and End-to-End (E2E) Latency The control loop is periodic with period P and deadline D≤PD≤ P. Under placement π, the nominal E2E latency is: L(π)≈∑v∈cv(π(v))+∑(vi,vj)∈ℰGdij(π(vi),π(vj)).L(π)≈ _v c_v(π(v))+\!\! _(v_i,v_j) _G\!\!d_ij(π(v_i),π(v_j)). (3) Denote that L is stochastic in practice due to workload variability and network jitter during operations; high-percentile values therefore serve as the primary QoS-aware target. I QoS-Aware Cost Model I-A Per-Window QoS Metrics The native DTP algorithm operates over observation windows of W control cycles indexed by k. Per single window, the original four QoS metrics are collected as follows: • L95,kL_95,k: Empirical 95th-percentile E2E latency, capturing tail behavior from network outliers and bursty compute; • VD,k∈[0,1]V_D,k∈[0,1]: Deadline violation rate (primary SLA); • Ur,kU_r,k, Ue,kU_e,k: CPU utilization across robot/edge nodes. Latency and utilization are normalized by design targets: L~95,k=L95,kL¯,U~r,k=Ur,kU¯r,U~e,k=Ue,kU¯e. L_95,k= L_95,k L, U_r,k= U_r,k U_r, U_e,k= U_e,k U_e. (4) VD,kV_D,k is bounded in [0,1][0,1] and requires no normalization. I-B Switching Penalty and Cost Function To penalize frequent migration, a switching penalty based on the Hamming distance between placements is applied: S(πk,πk−1)=1||∑v∈πk(v)≠πk−1(v).S( _k, _k-1)= 1|V| _v 1 \ _k(v)≠ _k-1(v) \. (5) This term suppresses “placement chatter”, high-frequency oscillation that degrades network stability and is the infrastructure-level manifestation of trust latency [1]. Given non-negative weights, the QoS-aware cost for window k is: Jk=αLL~95,k+αVVD,k+αrU~r,k+αeU~e,k+αSS(πk,πk−1).J_k= _L L_95,k+ _VV_D,k+ _r U_r,k+ _e U_e,k+ _SS( _k, _k-1). (6) The tension between αVVD,k _VV_D,k (deadline aggression) and αSS(πk,πk−1) _SS( _k, _k-1) (migration resistance) formalizes the core control–infrastructure trade-off. In industrial settings, αV>αL>αS _V> _L> _S reflects prioritizing SLA compliance as the primary, raw latency as the secondary, and stability as a soft constraint. I-C Constrained Placement Problem To keep per-window evaluation tractable on edge hardware, we originally restrict to a small interpretable candidate set Πcand _cand comprising the following three pilot placements: • LOC (fully local): T2T_2 on R1R_1, T3T_3 on R2R_2; • SO (static offload): T2T_2 and T3T_3 on E; • HYB (hybrid): T2T_2 on E, T3T_3 on R2R_2. The window-level placement problem is πk∗∈argminπ∈ΠcandJk(π;πk−1),π^*_k∈ _π∈ _candJ_k(π; _k-1), (7) subject to L95,k(π)≤LmaxL_95,k(π)≤ L_max and per-node utilisation caps Un,k(π)≤UmaxU_n,k(π)≤ U_max for all n∈n . IV Dynamic Task Placement Algorithm IV-A Window-Based DTP Procedure Algorithm 1 describes the DTP procedure running on the edge node – E. Two parameters jointly enforce hysteresis: • NminN_min: Minimum dwell windows; prevents rapid flapping when metrics fluctuate near thresholds; • Δmin _min: Cost-improvement threshold; ensures only statistically significant improvements trigger migration. Denote that per-window complexity is O(|Πcand|)O(| _cand|), which is always lightweight on any edge node if task placements occur. Algorithm 1 Window-Based Dynamic Task Placement (DTP) 0: W; (αL,αV,αr,αe,αS)( _L, _V, _r, _e, _S); Δmin _min; NminN_min; Πcand _cand 1: Init π0∈Πcand _0∈ _cand; dwell nd←0n_d← 0 2: for k=1,2,…k=1,2,… do 3: Apply πk−1 _k-1 for W cycles; collect L95,k,VD,k,Ur,k,Ue,k\L_95,k,V_D,k,U_r,k,U_e,k\ 4: Compute Jk(πk−1;πk−2)J_k( _k-1; _k-2) via (6) 5: if nd<Nminn_d<N_min then 6: nd←nd+1n_d← n_d+1; πk←πk−1 _k← _k-1; continue enforce min. dwell 7: end if 8: Estimate Jk(π;πk−1)J_k(π; _k-1) for all π∈Πcandπ∈ _cand 9: πk∗←argminπJk(π;πk−1)π^*_k← _π\,J_k(π; _k-1) 10: ΔJ←Jk(πk−1;πk−2)−Jk(πk∗;πk−1) J← J_k( _k-1; _k-2)-J_k(π^*_k; _k-1) 11: if ΔJ>Δmin J> _min then 12: πk←πk∗ _k←π^*_k; nd←0n_d← 0 migrate 13: else 14: πk←πk−1 _k← _k-1; nd←nd+1n_d← n_d+1 hold 15: end if 16: end for IV-B Estimating QoS for Non-Active Placements Scoring placements not currently active requires estimating their idle QoS metrics. The three initial complementary mechanisms are supported along the operation loop as follows: • Static profiling: Compute times cv(n)c_v(n) and communication delays measured offline and stored as lookup tables. • Online emulation: A small fraction of cycles is routed through each alternative in the background to update estimates continuously during the non-placement stage. • Conservative scaling: Upper-bound estimates derived from current measurements or passive monitored feedback and known compute/network ratios. The original algorithm is agnostic to the choice, provided cost comparisons remain qualitatively consistent for DTP. V Qualitative Analysis and Deployment Context V-A Anticipated Placement Trajectories Table I summarizes the expected steady-state DTP behaviors relative to LOC and SO baselines across four pilot canonical stress scenarios. The policy converges toward LOC under adverse network conditions and toward SO under robot CPU stress; HYB dominates when compute and communication resources are jointly constrained, by offloading the heavier T2T_2 (Perception) to E while keeping T3T_3 (Planning) on R2R_2. V-B Co-Design Implications and the PAI Nervous System Cast as the CFA 3C co-design principle-based orchestration layer, the DAG-based QoS-aware DTP framework provides three concrete guarantees against the failure modes intrinsic to networked factory control following industrial standards: • The hard-anchoring of T1T_1 and T4T_4 preserves the sub-millisecond reflex loop on local hardware, decoupling actuation safety and dependencies from network state; • The relocatable adaptation space accommodates the heterogeneous compute demands of T2T_2 and T3T_3, including emerging VLA/LLM workloads, without overloading machine hardware or hard-coding offload decisions; • The Hamming-distance switching penalty in (5) provides a mathematically bounded guarantee against placement chatter, translating directly to operator-observable control-loop stability and to the deterministic timing budgets expected of TSN-class substrates [9]. Technically, a factory automation system that cannot guarantee VD≤5%V_D≤ 5\% across the deployment envelope cannot credibly claim Industry 4.0 production maturity [1] toward industrial agentic automation for Industry 5.0 readiness. VI Empirical Strategy and Validation Roadmap Validation proceeds in two main phases, each with concrete parameters drawn directly from the physical robotic testbed to ensure direct comparability and reflection on the previous empirical LOC, SO, and ATP baselines. VI-A Phase 1 – Discrete-Event Simulation A discrete-event simulator parameterized will validate cost-function sensitivity and hysteresis bounds against the predictions in Table I. Key parameters are: • Control period P∈[20,50]P∈[20,50] ms with deadline D≤PD≤ P; • Baseline link RTT of 1–2 ms representative of an engineered industrial LAN, with TSN-class jitter envelopes as a sensitivity axis following standardization [9]; • CPU stress profiles applied to the R1R_1-equivalent node; • Gaussian network fault injection (e.g., μ=25μ=25 ms, σ=5σ=5 ms, ploss=2%p_loss=2\%), plus additional extended verification. VI-B Phase 2 – Hardware-in-the-Loop (HIL) The modular cloud-edge-robotics testbed will provide: • Measured compute profiles cv(n)c_v(n) on each physical node; • Empirical network delay distributions on the R1R_1–E and E–R2R_2 links with extended IIoT devices/PLCs; • Empirical CDFs of L95L_95, VDV_D, UrU_r, and UeU_e for LOC, SO, ATP, and DAG-QoS DTP across all stress scenarios. Phase 2 enables direct quantitative comparison between the proposed multi-stage DTP and the single-stage ATP baseline. VII Conclusions and Future Work In this WiP paper, we presented the DAG-based QoS-aware DTP, the formal multi-objective extension of our edge-based QoS-aware ATP framework. By formalizing the visual-servoing pipeline as the DAG, introducing the window-based cost function with the Hamming-distance switching penalty, and generalizing the ATP algorithm to concurrent multi-task placement, the extended framework addresses three structural limitations of single-task adaptive placement: limited scope, threshold-based decisions, and the absence of a formal switching cost model. The result is a runtime determinism layer compatible with the CFA-based 3C co-design ethos central to factory communications and mature industrial automation. Beyond factory robotic automation, this middle layer will act as one of the key cores in termed the “PAIs Nervous System”: the runtime infrastructure that lets LLM agents and VLA models operate reliably under real compute and network variability on different PAIs and nearby edges/clouds. Future work will extend the proposed framework in this WiP towards full deployment through: (i) empirical validation on the physical testbed; (i) expansion of the candidate placement set beyond LOC/SO/HYB and extension to multi-robot fleets where a single edge serves several DTP loops concurrently; (i) online cost-weight adaptation of hyperparameters via Bayesian optimization or lightweight meta-learning; (iv) joint compute–network co-scheduling with deterministic substrates, elevating the network from a delay distribution to a co-optimized hardware resource, and considering new QoS dimensions such as Age-of-Information (AoI). TABLE I: Expected dominant DTP placement per stress scenario. Scenario LOC SO DTP (expected) Baseline Low VDV_D; high UrU_r Low UrU_r; moderate VDV_D Mostly LOC; occasional HYB Robot CPU stress High VDV_D (UrU_r sat.) Mitigated; higher L95L_95 Converges to SO Edge CPU stress Stable; UrU_r elevated High VDV_D (UeU_e sat.) LOC or HYB Network impairment Low variance High VDV_D via jitter Converges to LOC References [1] R. Andreoli, R. Mini, P. Skarin, H. Gustafsson, J. Harmatos, L. Abeni, and T. Cucinotta (2025) A multi-domain survey on time-criticality in cloud computing. IEEE Trans. Serv. Comput. 18 (2), p. 1152–1170. External Links: Document Cited by: §I-B, §V-B. [2] J. Jin, Z. Pang, J. Kua, Q. Zhu, K. H. Johansson, N. Marchenko, and D. Cavalcanti (2025) Cloud-fog automation: The new paradigm towards autonomous industrial cyber-physical systems. IEEE J. Sel. Areas Commun.. Cited by: §I. [3] J. Jin, K. Yu, J. Kua, N. Zhang, Z. Pang, and Q. Han (2023) Cloud-fog automation: Vision, enabling technologies, and future research Directions. IEEE Trans. Ind. Inform. (), p. 1–16. External Links: Document Cited by: §I. [4] H. Li, X. Cao, Z. Zeng, Y. Wu, Y. Zhang, and Y. Xia (2026) OpenGo: An OpenClaw-based robotic dog with real-time skill switching. External Links: 2604.01708 Cited by: §I. [5] R. Li, Y. Zhou, Y. Zhu, K. Chen, J. Wang, S. Wang, K. Hu, M. Yu, B. Jiang, Z. Su, J. Ma, X. He, Y. Shen, Y. Yang, G. Ren, M. Yao, W. Wang, and Y. Mu (2026) RoboClaw: An agentic framework for scalable long-horizon robotic tasks. External Links: 2603.11558 Cited by: §I. [6] J. Liu, J. Ren, Y. Zhang, X. Peng, Y. Zhang, and Y. Yang (2023) Efficient dependent task offloading for multiple applications in MEC-cloud system. IEEE Trans. Mobile Comput. 22 (4), p. 2147–2162. External Links: Document Cited by: §I. [7] T. Luu, Q. Nguyen, T. Tran, M. Tran, S. Ding, J. Kua, and T. Hoang (2025-08) Enhancing real-time robot teleoperation with immersive virtual reality in industrial IoT networks. Int. J. Adv. Manuf. Technol. 139 (11), p. 6233–6257. External Links: ISSN 1433-3015, Document Cited by: §I. [8] Z. Tong, J. Deng, J. Mei, Y. Zhang, and K. Li (2024) Multi-objective DAG task offloading in MEC environment based on federated DQN with automated hyperparameter optimization. IEEE Trans. Serv. Comput. 17 (6), p. 3999–4012. External Links: Document Cited by: §I. [9] T. Zhang, G. Wang, C. Xue, J. Wang, M. Nixon, and S. Han (2024-10) Time-Sensitive Networking (TSN) for industrial automation: Current advances and Future directions. ACM Comput. Surv. 57 (2). External Links: ISSN 0360-0300, Document Cited by: §I, §I-A, 3rd item, 2nd item.