Non-Invasive Inspection of Water Canals Using Dronar

arXiv:2609.40178 · cs.RO, eess.SP · Submitted 2026-09-30 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Non-Invasive Inspection of Water Canals Using Dronar".

Dev: Open concrete canals play a vital role in water transportation, serving as primary water infrastructure for millions of people across the Phoenix, Arizona metro area.

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

Title and authors: Rosa: So we're looking at this paper titled "Non-Invasive Inspection of Water Canals Using Dronar," which tackles a big problem in infrastructure maintenance. It seems like they’ve developed a way to check concrete canals without having to drain the water, which is huge for operations.

Dev: That’s right, Rosa; the core idea is using affordable, off-the-shelf drone and sonar technology—which they call dronar—to inspect canal beds while keeping the water in there. It’s smart because it cuts out that four-year dry-up cycle that maintenance crews currently have to wait for before they can do any real inspection work.

Taro: I'm interested in how this bypasses the existing workflow constraints, Dev; if you can see issues without draining the water, that immediately changes how maintenance crews prioritize their efforts across a whole system.

Rosa: Exactly; it shifts the focus from reacting to dry-up events to proactively finding spots for targeted repairs right now. The paper’s summary outlines how this method helps locate both sediment buildup and potential lining deformation or concrete cracks before they become major problems, which is really important for scheduling.

Dev: From an engineering standpoint, the summary highlights that the main benefit is prioritizing maintenance crews based on where the sediment buildup is located so they can spend their time efficiently. It’s about making inspections actionable immediately instead of waiting years for a dry-up.

Taro: But I want to know what happens when things go wrong; if you have this system operating autonomously in real canal conditions, what kind of unexpected issues might the AI need to handle?

Rosa: That leads us perfectly into the improvements section; the authors are already suggesting ways to make this tool even more powerful by integrating multimodal data and developing sophisticated detection models. They aren't just stopping at measuring depth; they’re looking at using that sonar data with imagery.

Dev: That sounds like a significant step up from just getting a simple depth profile, Rosa; fusing the DownScan data with what they can get from SideScan imagery could give them much richer information about the physical structure of the canal bed itself.

Title and authors: Taro: If you can correlate those sonar measurements with visual evidence of cracks or deformations, that moves it beyond simple sediment detection and into identifying structural integrity issues, which is where autonomy really matters when things are misbehaving.

Rosa: Right; the suggestion there is to build a real-time anomaly detection model that classifies conditions like clean versus sedimented based on those continuous depth profiles they’re collecting. It sounds like they want the system to tell them exactly what’s wrong right as it sees it happening in the water.

Dev: And from my side, I'm thinking about how reliable that real-time classification needs to be; we need a model that can handle variations in flow or even temperature changes without giving false alarms, which is a tough constraint for any deployed system.

Taro: That’s the challenge of autonomy in the field; if the world throws unexpected variables at your sensor, your system needs to adapt its strategy dynamically rather than just following a pre-programmed path.

Rosa: The paper also suggests an autonomous path planning approach where the USV can adjust its survey pattern based on real-time sensor feedback and flow dynamics, which is crucial for navigating those moving water environments effectively.

Dev: I worry about the loop rate there; if you’re constantly recalculating your patrol route based on immediate feedback, that introduces latency into every decision, and we need to make sure that doesn't create a dangerous lag in response.

Taro: That’s where reinforcement learning could really help optimize the route for maximum coverage of high-risk areas while making smart trade-offs between speed and energy usage.

Rosa: So, the authors are proposing a way to use reinforcement learning to make the survey strategy itself adaptive, optimizing coverage based on what it finds along the way. It’s about making the inspection smarter as it goes.

Dev: That makes sense in theory, but I need assurance that this agent can handle failure modes gracefully; if its path planning gets stuck or loses tracking suddenly, we need robust fail-safes built into the control loop.

Taro: And when we think about those larger implications, this kind of non-invasive inspection capability means infrastructure management could become vastly more proactive instead of reactive, which is a big deal for public safety in areas like the Phoenix metro.

Title and authors: Rosa: I agree; having this technology scalable for a large network means we could start catching issues way earlier than waiting for those dry-up cycles, potentially preventing serious leaks or structural failures down the road.

Dev: But the practical deployment requires us to consider how long this system can actually run on battery before it needs intervention, because continuous operation is key for thorough inspection.

Taro: That points toward needing robust power management integrated into the autonomy framework so that long-duration surveys are feasible without constant recharging interruptions.

Rosa: So, to wrap up on these improvements, the paper pushes for a fully automated post-processing pipeline that can handle those complex filtering steps themselves, which would drastically cut down the manual analysis time for maintenance crews.

Dev: Automating that filtering is a smart move because it removes human error from the data cleaning process; if we can automate isolating those intermediate-scale variations, the resulting reports will be much more consistent.

Taro: That level of automation in data processing is what makes a system truly scalable for widespread use; it moves the bottleneck from human analysis to machine execution.

Rosa: In conclusion, this paper on "Non-Invasive Inspection of Water Canals Using Dronar" shows a viable proof-of-concept for using accessible technology to inspect canal beds without draining the water, proving we can get repeatable measurements under operational conditions.

Dev: It’s definitely a solid foundation because they’ve demonstrated repeatability, showing that the system produces depth profiles that overlap closely when run repeatedly along a clean segment.

Taro: For me, the implication is that this opens up a whole new class of monitoring tools for water infrastructure across various regions, not just the SRP system mentioned in their tests.

Rosa: That's what I think; it’s about creating a scalable tool that can be adapted to many different types of canal networks because it relies on modular, off-the-shelf components.

Dev: We just need to keep focusing on those technical hurdles regarding latency and ensuring the system doesn't fail when the real world throws unexpected turbulence at it during deployment.

Taro: And looking ahead, I see this leading toward systems that can handle more complex structural defects like lining deformation, which is where the next level of autonomy will really need to focus its attention.

The paper's summary: Rosa: So, to wrap up on this paper on dronar, they’ve shown that by putting consumer-grade sonar into a drone and surface vehicle, you can check canal beds without draining the water for those four years of dry-up cycles.

Dev: That's the core idea; it’s about using existing tools to create a non-invasive inspection method that can be deployed immediately. It really addresses the current bottleneck where maintenance has to wait for specific weather conditions just to look at things.

Taro: And what I find interesting is how they're framing this as a proof-of-concept tool, which suggests it’s not just some lab experiment but something that could actually be used in the field for routine checks.

Rosa: Exactly; the summary emphasizes that this system has already proven it can detect sediment buildup and produce repeatable depth measurements even when there's water moving around. They showed you can compare a segment that got cleaned against one that hasn't, and the differences are clearly visible on the sonar data.

Dev: The repeatability they achieved, with those seven-centimeter overlaps in the depth profiles, is what really convinces me about its reliability under operational canal conditions; it’s not just a one-off measurement.

Taro: That repeatable data is crucial because it moves this from a proof-of-concept into something that could actually be used to establish baseline conditions for monitoring degradation over time.

Rosa: Right, and the authors conclude that this dronar setup is a viable starting point for a scalable system because it relies on affordable hardware that maintenance crews can access and use right away.

Dev: I agree; the modular nature of the components makes it much more accessible than developing some highly specialized, expensive equipment for every single project.

Taro: Thinking about the larger impact, if this technique scales across a whole network of canals, it could fundamentally change how we monitor aging infrastructure before major failures occur.

Rosa: It really does; instead of waiting for catastrophic issues to appear during a dry-up period, you could have routine inspections happening continuously or on a much more frequent schedule.

Dev: That proactive approach is what keeps me focused on the technical side—we need to make sure the system can handle the real-world noise and turbulence without breaking down under pressure.

Taro: And that’s exactly where I want to look next; if we can build on this repeatability with more sophisticated analysis, we could start looking at identifying other types of damage, like those lining deformations they mentioned as a future goal.

The paper's improvements: Taro: So, we've covered how dronar works for basic sediment detection; now let's talk about the suggested improvements to make this tool even more capable of handling complex infrastructure issues and autonomous decision-making when things get messy.

Rosa: The paper suggests a few key upgrades, starting with building a real-time anomaly detection model that can classify canal conditions, meaning the AI can automatically tell you if there's just sediment or if there's actually some lining deformation present.

Dev: That’s smart because it moves beyond simple depth measurement; we need the system to differentiate between normal variations and actual structural problems, which requires a supervised machine learning model trained on those pre- and post-maintenance profiles they tested.

Taro: And I think the next big step is integrating semantic feature identification to localize those defects spatially, correlating the sonar data with imagery to pinpoint exactly where a crack or deformation is occurring.

Rosa: I love that idea of fusing DownScan data with SideScan imagery; that combination should give us a much richer picture of the canal's internal structure than just looking at depth alone.

Dev: From my side, I’m concerned about how this model handles those real-time inputs; we need to make sure the classification is fast enough for operational use without introducing significant latency into the control loop.

Taro: And that brings up the autonomous path planning suggestion, where a reinforcement learning agent adjusts the USV's survey strategy based on flow dynamics and energy levels, optimizing coverage in real-time.

Rosa: That’s what I’m excited about; it means the system won't just follow a fixed route but will actually adapt its patrol based on what it discovers along the way, which is essential for efficiency.

Dev: While dynamic path planning sounds great, I want to see how robust that agent is when the world misbehaves unexpectedly; we need solid fail-safes built into that decision-making process so it doesn't get stuck or make a bad move.

Taro: That’s where we can look at applying concepts from papers like RoboHarness, which deals with orchestrating heterogeneous policies for long-horizon tasks, to build that adaptive agent.

Rosa: And finally, they propose automating a post-processing pipeline to handle the filtering steps themselves; this would cut down on the manual analysis time considerably and make generating those standardized reports much faster for maintenance crews.

Dev: Automating that filtering is a solid move because it removes human error from cleaning the data, provided we can get that bandpass filtering algorithm to work consistently in varied water conditions.

Taro: If we can automate that complex data processing, it means the system becomes truly self-sufficient for initial assessment, which is a big step toward making it truly autonomous.

Rosa: So these improvements really push dronar from being a simple measurement tool into something that could be a full-fledged monitoring system capable of identifying structural issues and making smarter operational choices on its own.

Conclusion: Rosa: So, to wrap up on the paper "Non-Invasive Inspection of Water Canals Using Dronar," they’ve clearly established that this drone and sonar integration is a viable way to inspect water infrastructure without having to drain it for months.

Dev: That’s right; the results showed repeatable depth profiles, which means we can trust the data collected under actual operational flow conditions rather than just static lab tests.

Taro: And if we look at the future work mentioned, it really points toward integrating structural defect identification alongside that monitoring capability to see what else this system can find down there.

Rosa: Exactly; they're moving from just detecting sediment to actually mapping out physical damage like lining issues, which is a significant step for proactive maintenance planning.

Dev: From an engineering standpoint, the authors flag that the current method doesn't cover every potential defect yet, so we need to focus on how the AI can be extended to handle those more complex structural anomalies we discussed earlier.

Taro: I agree; if we can get that classification model running reliably in the field, it opens up possibilities for using these drones across a whole network for routine health checks.

Rosa: It really does; this system could drastically reduce the downtime and cost associated with infrastructure repairs by catching problems way earlier than waiting for seasonal dry-down cycles.

Dev: I’m just thinking about how we manage that real-time data stream over a long survey; if we can nail the loop rate and control latency, this becomes a practical inspection tool instead of just an interesting proof-of-concept.

Taro: That brings up the autonomy challenge again; as the system gets smarter, how do we ensure it doesn't make incorrect decisions when it encounters unexpected turbulence or sensor noise in a real canal environment?

Rosa: We’ll have to work on those safety parameters closely; that’s where field testing really comes into play to see how long and under what conditions this setup can actually perform reliably.

Dev: So, the paper lays a solid foundation for using modular, accessible hardware for infrastructure monitoring while still acknowledging the need for deeper integration and more robust autonomous decision-making.

Taro: It certainly sets a high bar for what’s possible when we combine consumer tech with advanced autonomy; this kind of approach is going to influence how we look at inspection tools in other complex environments.

Rosa: We’ve got a lot to chew on with this paper, but next time, we'll be looking at papers that push the boundaries of robotic control and how AI handles those messy real-world interactions.

Michael Zielinski, Zhizhan Wang, Benjamin Z. Dymond, Ph.D., Reza Razavian, Ph.D., Zhongwang Dou, Ph.D.

Mechanical Engineering Department, Steve Sanghi College of Engineering, Northern Arizona University · Civil and Environmental Engineering Department, Steve Sanghi College of Engineering, Northern Arizona University · Department of Biomedical Engineering, Kate Gleason College of Engineering, Rochester Institute of Technology

cs.RO, eess.SP

Submitted: 2026-09-30

Updated: 2026-09-30

Code: https://github.com/ArduPilot/ardupilot

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 73/100

The gist: Open concrete canals play a vital role in water transportation, serving as primary water infrastructure for millions of people across the Phoenix, Arizona metro area.

Key concepts

Dronar System: A drone-based inspection system that uses an unmanned surface vehicle (USV) to travel through canals and collect sonar data. It combines consumer-grade sonar with a custom twin-pontoon USV platform, allowing for non-invasive measurement of canal conditions.
Sonar Transducer: The Active Imaging HD 3-in-1 transducer paired with an HDS Pro 9 sonar hub. This component operates at 800 kHz and provides both DownScan coverage (measuring depth from above) up to ±30 degrees and SideScan coverage up to ±58 degrees, allowing for detailed imaging of the canal bed.
Repeatability Analysis: A method used to verify the system's accuracy by comparing multiple runs along a segment. By filtering out noise and isolating variations, the study confirmed that repeated surveys produced depth profiles that overlapped closely, proving the system can produce consistent measurements under operational conditions.

Terminology

Core — 3× in this paper · explained in 2 episodes
Reinforcement Learning — 3× in this paper · explained in 4 episodes
Salt River Project — 3× in this paper
Trace — 3× in this paper · explained in 2 episodes
Accuracy — 2× in this paper · explained in 1 episode
Consistency — 1× in this paper · explained in 3 episodes
Failure Modes — 1× in this paper · explained in 3 episodes
Structural Integrity — 1× in this paper · explained in 3 episodes
Verification — 1× in this paper · explained in 2 episodes

Summary

Open concrete canals play a vital role in water transportation, serving as primary water infrastructure for millions of people across the Phoenix, Arizona metro area. The research team developed and verified an easily deployable and non-invasive method to inspect canal beds without draining the water by integrating affordable, off-the-shelf drone and sonar technology (termed dronar). This system addresses the resource-intensive process of identifying issues like lining deformation or sediment buildup that currently only occurs during four-year dry-up cycles, establishing a viable proof-of-concept tool for routine canal bed inspection.

Dronar System Integration

The dronar system integrates consumer-grade sonar technology into an unmanned surface vehicle (USV) to traverse the length of the canals and collect sonar data. The specific components include:

A sonar transducer: An Active Imaging HD 3-in-1 transducer paired with an HDS Pro 9 sonar hub by Lowrance Electronics, which operates at 800 kHz and provides DownScan coverage up to ±30◦ and SideScan coverage of up to ±58◦.
The USV platform: A custom-built twin-pontoon surface vehicle consisting of two kayak stabilizer pontoons (Brocraft Kayak Outrigger) and an aluminum frame, propelled by differential thrust from two electric thrusters (ApisQueen U2).
Control and telemetry: A CubePilot Cube Orange+ mounted on a Kore carrier board, programmed to run the rover firmware in ArduPilot. Remote control is maintained via a Herelink Air Unit with extended-range antennas transmitting location coordinates (Here4 GPS module).

Operational Deployment and Recovery

The deployment and recovery of the dronar system were addressed by developing a dual-purpose transport and deployment cart featuring custom aluminum rails and a rear-mounted retention hook. This cart facilitates rope-guided descent/ascent of the steep canal walls (45◦–60◦) during launch and recovery. The system was designed to be fully untethered, with no physical cables between the vehicle and the operator, relying on local storage for all sonar data and controller variables.

Field Testing Methodology

The research conducted three field tests across the Arizona Canal in Phoenix, AZ. Test 1 served as a proof-of-concept survey on a recently cleaned segment with limited sediment buildup to assess vehicle performance in moving water and evaluate the cart-based launch and recovery method. Test 2 focused on verifying repeatability by surveying a segment that experienced heavy sedimentation, comparing pre- and post-maintenance data. For Test 2b (post-maintenance), six downstream-only runs were performed, including Manual mode at full throttle, Manual mode at canal water current speed (with side-to-side corrective inputs only), and Auto mode.

Sonar Data Analysis and Repeatability

The primary analysis focused on DownScan depth profiles to detect sediment buildup. The study utilized a bandpass filter applied to the depth profiles to "isolate intermediate-scale bed variations: spatial variations at wavelengths longer than 50 m were suppressed to remove the background depth trend, while variations shorter than 0.05 m were suppressed to remove high-frequency measurement noise. This filtering allowed for comparisons where the sedimented and cleaned segments [were] placed on a common reference and making differences in canal bed condition directly visible. The results confirmed repeatability: Repeated runs along the clean segment produced closely overlapping depth profiles, with measurements tracked within a maximum of 7 cm of each other. Furthermore, comparing pre-maintenance (Test 2a) and post-maintenance (Test 2b) raw water depth profiles showed that the pre-maintenance trace exhibited pronounced negative residuals at multiple locations along the segment, reaching as large as 15 cm," which correspond to locations where sediment accumulation significantly reduced the sonar-measured depth below the local mean.

Conclusion on Feasibility

The study concluded that the dronar system is a viable proof-of-concept tool for non-invasive canal bed inspection. The system successfully demonstrated its capability to detect sediment accumulation and produce repeatable measurements under operational canal conditions. The findings establish the dronar’s potential as a scalable system for use during routine inspection across a large concrete water canal network due to its use of affordable, modular hardware and ability to operate in active canals (without draining the water for inspection). Limitations identified include the need for validation against other defects like lining deformation and future development of automated post-processing pipelines.

The gist: The dronar system reliably detects sediment accumulation in water canals by producing repeatable depth profiles under operational conditions, positioning it as a scalable tool for non-invasive infrastructure inspection.

How it works

The core mechanism relies on the integration of consumer-grade sonar and drone technology onto an autonomous unmanned surface vehicle (USV). The USV is equipped with an Active Imaging HD 3-in-1 transducer that provides both DownScan and SideScan coverage.

Improvements for AI systems

Here are specific improvements to AI systems based on the principles and capabilities demonstrated by the Dronar system described in this paper:

The core improvement lies in developing a multimodal, autonomous inspection framework that integrates aerial (drone) and surface (USV) data streams for real-time infrastructure assessment.

// Improvement: Development of a Real-Time Anomaly Detection Model for Canal Bed Condition Classification
// Capability: The AI system can ingest continuous, time-series DownScan sonar depth profiles (as demonstrated in Figure 11) and classify the canal bed condition (e.g., clean vs. sedimented) with high precision, even under varying flow conditions or ambient temperature influences (addressing thermal management observations).
// Specific Function: Implement a supervised machine learning model (e.g., a Convolutional Neural Network or Recurrent Neural Network) trained on the pre- and post-maintenance depth profiles to automatically flag areas exhibiting significant negative residuals (sediment accumulation) exceeding a predefined threshold (e.g., >10 cm reduction in depth).
// Improvement: Integration of Semantic Feature Identification for Defect Localization
// Capability: The AI can correlate detected sediment deposits with structural anomalies by fusing DownScan data with SideScan imagery (as collected but not analyzed) to identify subtle linear features indicative of canal lining deformation or concrete cracks, moving beyond simple depth measurement.
// Specific Function: Develop an object detection and segmentation pipeline that, when paired with a known canal geometry model, can automatically delineate the precise spatial extent and character (e.g., crack width estimate) of deformations detected in the imagery.
// Improvement: Autonomous Path Planning and Adaptive Survey Strategy
// Capability: The AI system can dynamically adjust the USV's operational mode (Manual vs. Auto) and survey pattern based on real-time sensor feedback, flow dynamics, and battery status (addressing speed/endurance observations).
// Specific Function: Implement a reinforcement learning agent that optimizes the patrol route to maximize coverage of high-risk areas identified by the initial sonar sweep while minimizing energy expenditure, transitioning seamlessly between downstream and upstream survey strategies based on current conditions.
// Improvement: Automated Data Post-Processing Pipeline (Semi-Automated)
// Capability: The system can automate the manual post-processing steps currently performed using Python scripts (Martinsen 2023), specifically focusing on filtering depth profiles to isolate intermediate-scale bed variations and generating standardized reports for maintenance crews.
// Specific Function: Deploy an automated bandpass filtering algorithm that isolates the relevant spatial features (e.g., sediment deposits) from background noise (canal grade/surface slope), resulting in a quantifiable metric of canal degradation, significantly reducing human analysis time from hours to seconds per segment.

Abstract

Open concrete canals play a vital role in water transportation, serving as primary water infrastructure for millions of people across the Phoenix, Arizona, metro area. Over time, the concrete canals can experience a range of issues, including canal lining deformation, cracked concrete, and sediment buildup on the canal floor. Identifying such critical issues is a resource-intensive process, which currently happens only during four-year dry-up cycles. This prevents the maintenance crew from prioritizing operations on the most affected canal segments. To address this issue, the research team has developed and verified an easily deployable and non-invasive method to inspect canal beds without draining the water. This inspection system integrates affordable, off-the-shelf drone and sonar technology (termed dronar). This dronar system includes a consumer-grade sonar system integrated into an unmanned surface vehicle (USV) that carries the sonar transducer just under the surface of the canal water. This paper presents a proof-of-concept demonstration of the dronar system across three field tests on the Arizona Canal in Phoenix. DownScan depth profiles from the sedimented canal segment were consistently shallower than profiles from the same segment after cleaning, with offsets of up to 15 cm observed along the track. Repeated runs over the clean segment produced closely overlapping DownScan depth profiles, confirming that the dronar yields repeatable measurements across the natural variation of the canal bed. These results establish the dronar as a viable proof-of-concept tool for non-invasive canal bed inspection.