FLoRa: Flight-Assisted Data Collection from Duty Cycling LoRa Nodes under Energy Constraints

arXiv:2610.09226 · cs.NI, cs.LG · Submitted 2026-10-06 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: Today's paper: "FLoRa: Flight-Assisted Data Collection from Duty Cycling LoRa Nodes under Energy Constraints".

Jane: The gist The proposed FLoRa framework is a hierarchical optimization architecture using Simulated Annealing for path planning, Covariance Matrix Adaptation Evolution Strategy for hover positioning,

Tom: First, who's behind it and why it matters.

Title and authors: Tom: So we're looking at this paper now, FLoRa: Flight-Assisted Data Collection from Duty Cycling LoRa Nodes under Energy Constraints. It sounds really specific about solving a tough problem with duty-cycling LoRa devices and battery limits.

Jane: Right, it’s about making sure a drone collects data from these low-power nodes without running out of juice. The authors, Kakarla and Mahendran, they are tackling this by proposing a multi-level optimization architecture to handle the routing, the hovering position, and the probing decisions all at once.

Lu: It's interesting because it’s structured as a hierarchy where you have discrete combinatorial optimization for routing using Simulated Annealing, then continuous global optimization for spatial positioning using Covariance Matrix Adaptation Evolution Strategy, and finally sequential decision-making under uncertainty with Partially Observable Markov Decision Processes.

Meng: A hierarchical approach sounds promising on paper, but from my side, I wonder if the complexity of chaining those three things together will make it too slow to actually deploy in the field.

Lalam: It's aiming to solve the Probabilistic, Timedependent, and Close-enough Orienteering Problem under energy constraints <ref:2610.09226#pg1>.

The paper's summary: Jane: So what’s the core idea behind FLoRa? They frame it as a flight-assisted LoRa data collection architecture designed to solve that Orienteering Problem we talked about earlier, focusing on maximizing the total expected VIP freshness of the collected data while staying within those hard energy budget constraints.

Tom: That VIP metric is a key thing they introduce because it rewards fresh data and penalizes failed probes, which makes sure the problem is well-posed especially when a LoRa node isn't available <ref:2610.09226#pg1>.

Lu: The paper lays out this hierarchy clearly: first, Simulated Annealing explores candidate UAV tours to decide the order of visits and subset selection under those hard flight energy budget constraints <ref:2610.09226#pg3>. Then, for each tour, Covariance Matrix Adaptation Evolution Strategy finds the best hovering position within the coverage neighborhood of every visited LoRa node <ref:2610.09226#pg3>.

Meng: Okay, so they figure out the path first, then they optimize where to hover at each stop. But what about when you're actually there? The probing part seems to be handled by something else entirely.

Jane: Exactly, that’s where the third layer comes in: a network-wide POMDP for probing IoTDs <ref:2610.09226#pg3>. This POMDP helps the UAV make sequential decisions about whether to probe a node or fly to the next one given the uncertainty about when that node will be available.

Lalam: They manage the complexity of that network-wide POMDP by using surrogate methods like Lagrangian relaxation and forward decomposition to approximate inter-node time dependency <ref:2610.09226#pg3>.

The paper's improvements: Tom: The authors really focused on making this tractable, because solving the Mixed-Integer Non-Linear Programming problem directly is just not feasible <ref:2610.09226#pg7>. They use a nested decomposition approach where each level optimizes one decision variable while treating the others as fixed parameters.

Jane: That’s smart because it breaks down the massive problem into smaller, manageable pieces: tour selection, hover position, and probing decision. They also tackled the battery constraints by relaxing them into soft average constraints using Lagrangian relaxation <ref:2610.09226#pg3>.

Lu: And they approximated the inter-node time dependency using a forward decomposition technique to decouple that global shared time resource <ref:2610.09226#pg3>. On the node level, they solve the POMDPs using threshold-based optimal policies derived from a dual problem solved by the bisection method for the Lagrange multiplier lambda <ref:2610.09226#pg1>.

Meng: So it's a lot of mathematical machinery just to make sure it runs. From an engineering standpoint, I’m curious how much does this approximation actually affect the precision of the final path?

Jane: That’s a fair question, Meng. They show that this hierarchical structure acts as a low-pass filter which plays the role of a regularizer for SF-induced discontinuity <ref:2610.09226#pg3>. It seems to tame those hard discontinuities that happen when you consider signal strength regions, or SF regions.

Conclusion: Tom: So, in summary, FLoRa is the first hierarchical optimization framework they've proposed to jointly solve the Probabilistic, Timedependent, and Close-enough Orienteering Problem for UAV data collection under average energy budget constraints <ref:2610.09226#pg1>. It combines SA for planning, CMA-ES for positioning, and POMDPs for probing.

Jane: The results show it beats metaheuristic, greedy, and deep reinforcement learning baselines by significant margins—specifically thirty point six percent, twenty-seven point eight percent, and fifteen point two percent in total expected VIP across different metrics like coverage <ref:2610.09226#pg1>. It’s shown to achieve feasibility at low energy budgets where other methods fail for hover positioning <ref:2610.09226#pg1>.

Lu: From a research standpoint, the combination of SA, CMA-ES, and POMDP provides a robust solution that regularizes rigidities and discontinuities through Lagrangian relaxation and bi-level optimization <ref:2610.09226#pg13>. It's a solid structure for handling these kinds of problems.

Meng: For deployment, the key thing I see is that they managed to make it run without adding extra state augmentation that would make the curse of dimensionality worse <ref:2610.09226#pg3>. If you can deploy something that doesn't require massive state spaces, that’s a win for me.

Lalam: I’m just confirming what they flag as a limitation: the framework is designed to solve it under average energy budget constraints, and they use Lagrangian relaxation to manage those hard battery constraints <ref:2610.09226#pg3>.

Tom: So, we have this FLoRa framework that’s promising for real-world UAV data collection under duty-cycling LoRa conditions. That’s where we are with the paper on FLoRa: Flight-Assisted Data Collection from Duty Cycling LoRa Nodes under Energy Constraints.

Naresh Babu Kakarla, V. Mahendran

Indian Institute of Technology Tirupati

cs.NI, cs.LG

Submitted: 2026-10-06

Updated: 2026-10-06

Comments: 20 pages

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 78/100

The gist: The gist The proposed FLoRa framework is a hierarchical optimization architecture using Simulated Annealing for path planning, Covariance Matrix Adaptation Evolution Strategy for hover positioning,

Key concepts

Simulated Annealing (SA)
SA is used as the first layer of optimization to determine the best sequence of LoRa nodes a UAV should visit. It explores a large space of possible routes, effectively solving a complex routing problem by iteratively accepting or rejecting paths based on an energy budget constraint.
Covariance Matrix Adaptation Evolution Strategy (CMA-ES)
This strategy is employed to solve the 'close-enough' positioning problem. For a chosen route, CMA-ES calculates the best hovering location near each LoRa node that maximizes data collection efficiency within its local coverage area, optimizing continuous spatial decisions.
Partially Observable Markov Decision Processes (POMDP)
POMDPs manage the uncertainty of when to probe a LoRa node. This sequential decision-making process helps the UAV decide whether to attempt a data collection probe based on the probability of finding available nodes, considering energy costs and temporal dependencies.
Value of Information for Pull-based systems (VIP)
VIP is a metric used to evaluate data freshness. It rewards successful, timely collections while penalizing failed probes by measuring the time spent trying to connect. Maximizing total VIP freshness is the primary objective of FLoRa.

Summary

The gist The proposed FLoRa framework is a hierarchical optimization architecture using Simulated Annealing for path planning, Covariance Matrix Adaptation Evolution Strategy for hover positioning, and Partially Observable Markov Decision Processes for probing LoRa nodes under energy constraints.

FLoRa Framework Overview

FLoRa is introduced as a Flight-assisted LoRa data collection architecture designed to solve the Probabilistic, Timedependent, and Close-enough Orienteering Problem (PTCOP) (Page 1). This framework jointly optimizes four key aspects: discrete combinatorial optimization for routing, continuous global optimization for spatial positioning, and sequential decision-making under uncertainty (Page 1). The hierarchical structure is depicted in Fig. 2 and consists of three main layers: Simulated Annealing (SA) for path planning, Covariance Matrix Adaptation Evolution Strategy (CMA-ES) for hover positioning, and a Network-wide POMDP for probing IoTDs (Page 3).

Key Components and Metrics

The framework incorporates several novel elements to address the challenges of duty-cycling LoRa nodes (Page 1). A novel value-based freshness metric, Value of Information for Pull-based systems (VIP), is proposed, which rewards fresh data and penalizes failed probes imposing well-posedness (Page 1). The VIP metric is defined as Vi = (T − Pi j=1 τj if collected and Pi j=1 τj ≤ T, −τi otherwise) where τi penalizes a failed attempt by the time spent at that node (Page 4). The primary objective of FLoRa is To maximize the total expected VIP freshness of the collected data subject to the hard energy budget constraints of the UAV (Page 7).

Hierarchical Solution Structure

The problem is formulated as a Mixed-Integer Non-Linear Programming (MINLP) problem that is computationally intractable to solve directly (Page 7). To tackle this, FLoRa employs a nested decomposition approach where each level optimizes one decision variable while treating the others as fixed parameters (Page 7). The optimization structure is defined by the following subproblems:

  1. Tour selection z = (zi) z i=1 ∈ Z, optimized via Simulated Annealing (SA) to solve the Orienteering Problem (OP) under hard flight energy budget constraints (Page 8).

  2. Hover position l = (li) z i=1, optimized via CMA-ES to solve the Close-Enough problem in the coverage neighborhood of each IoTD for a given tour z (Page 9).

  3. Probing decision n = (ni) z i=1, optimized via POMDP to solve the temporal dilemma under probabilistic IoTD availability (Page 10).

Tractability through Approximation

The complexity of the network-wide POMDP is managed through several surrogate methods (Page 3). The hard battery constraints are relaxed into soft average constraints via Lagrangian relaxation, avoiding state augmentation that worsens the curse of dimensionality (Page 3). Furthermore, inter-node time dependency is approximated using a forward decomposition technique to decouple the global shared time resource (Page 3). The node-level POMDPs are solved using threshold-based optimal policies derived from a dual problem solved by the bisection method for the Lagrange multiplier λ (Page 10).

Performance and Evaluation

Evaluation shows that FLoRa outperforms metaheuristic, greedy, and deep reinforcement learning baselines by significant margins (Page 1). Specifically, FLoRa outperforms these baselines by 30.6%, 27.8%, and 15.2% in total expected VIP (Page 1). This advantage is demonstrated across various metrics: Total Expected VIP, Node Coverage, and Expected successful collections (Page 14). The close-enough positioning paradigm is shown to be critical at low energy budgets, where FLoRa achieves feasibility that above-node baselines cannot attain (Page 15).

Conclusion

The proposed FLoRa framework is the first hierarchical optimization framework to jointly solve the PTCOP problem setting for UAV data collection over LoRa nodes under average energy budget constraints (Page 2). The combination of SA, CMA-ES, and POMDP provides a robust solution that regularizes rigidities and discontinuities in the problem structure through Lagrangian relaxation and bi-level optimization (Page 13). Extensive simulation results confirm FLoRa's superior performance over existing methods across different battery budgets and network densities (Page 17).

--- Page 1 ---

IEEE INTERNET OF THINGS JOURNAL: UNDER REVIEW 1

FLoRa: Flight-Assisted Data Collection from Duty Cycling LoRa Nodes under Energy Constraints

Naresh Babu Kakarla and V. Mahendran (Page 1)

How it works

The hierarchical structure of the proposed FLoRa framework is depicted in Fig. 2 and its various layers are described as follows: Simulated Annealing (SA) [24], [25] typically used to solve NP-hard combinatorial problems is used to explore largesolution space of the path-planning aimed at deciding the order of visits to IoTDs (Page 3). Each candidate path (chosen by SA) is evaluated by invoking Covariance Matrix Adaptation Evolution Strategy (CMA-ES) [26] which in turn will address the close-enough problem of finding a suitable hovering position in the coverage neighborhood of each of the IoTD’s (Page 3). The top two layers of SA and CMA-ES consider hard (battery) constraints (without Lagrangian relaxation) (Page 3).

Hierarchical Solution Structure

The problem is formulated as a Mixed-Integer Non-Linear Programming (MINLP) problem that is computationally intractable to solve directly (Page 7). To tackle this, FLoRa employs a nested decomposition approach where each level optimizes one decision variable while treating the others as fixed parameters (Page 7). The optimization structure is defined by the following subproblems:

  1. Tour selection z = (zi) z i=1 ∈ Z, optimized via Simulated Annealing (SA) to solve the Orienteering Problem (OP) under hard flight energy budget constraints (Page 8).

  2. Hover position l = (li) z i=1, optimized via CMA-ES to solve the Close-Enough problem in the coverage neighborhood of each of the IoTD’s for a given tour z (Page 9).

  3. Probing decision n = (ni) z i=1, optimized via POMDP to solve the temporal dilemma under probabilistic IoTD availability (Page 10).

--- Page 2 ---

IEEE INTERNET OF THINGS JOURNAL: UNDER REVIEW 2

0 2000 4000 6000 8000 10,555,12,84 (Page numbering is based on the provided text structure)

How it works

The UAV consumes energy in two phases during data collection: flight between nodes and hovering at each node for data transfer (Page 6). The hovering power consumption h is computed as [42] (Page 6). The flight power consumption when the UAV travels at velocity v is [42] (Page 6).

Improvements for AI systems

  1. Bold tour selection via Simulated Annealing (SA) to maximize total expected VIP: The SA algorithm navigates into the large solution space effectively to solve the following path planning problem and optimizes "subset selection & visiting order" by maximizing the objective function defined in Eq. (11).

  2. Adaptive hover positioning via Covariance Matrix Adaptation Evolution Strategy (CMA-ES): CMA-ES is used to find the best UAV hovering positions within the coverage neighborhood of each of the IoTD’s by learning a narrow ridge shape and stretch/rotate it into a suitable ellipsoid to match the narrow ridges in Eq. (2).

  3. Probabilistic probing via Partially Observable Markov Decision Processes (POMDP): The system solves the temporal dilemma using POMDPs, where the UAV constructs a belief using the underlying IoTD’s data availability probability to determine the optimal number of probing attempts at each hovering location under random IoTD availability.

  4. Well-posed freshness metric via Value of Information for Pull-based systems (VIP): The proposed VIP metric rewards successful collection while penalizing failed probe attempts, ensuring the problem is well-posed especially when the IoTD is unavailable by preventing the UAV to probe infinitely.

  5. Tractable network optimization via Lagrangian relaxation and forward decomposition: To handle battery constraints and global time dependency, the framework uses these surrogates to solve node-level POMDPs, decomposing the network POMDP into node-level POMDPs by approximating inter-node time dependency using a forwarddecomposition technique.

  6. Dynamic probing policy via Threshold Policy: The optimal probing strategy is determined by a threshold policy, which dictates that the UAV should probe if and only if xt s li i ≥ t l i (1 + λh) / (T − Pi−1 j=1 Enj[τj] − ti−1,i before terminating.

Sources

Related papers