1. Problem Statement
The United States is spending $4.677 billion to plug orphaned oil and gas wells and cannot find most of them.
Section 40601 of the Infrastructure Investment and Jobs Act appropriated that sum across three programs: $250 million for wells on federal lands and waters, $4.275 billion for wells on state and private lands, and $150 million for wells on Tribal lands (U.S. Department of the Interior, Orphaned Wells Program FY 2025 Annual Report to Congress, November 2025). Through 30 September 2025 the Orphaned Wells Program Office had obligated roughly $1.8 billion, and states had plugged 10,257 wells.
The binding constraint is not funding and not plugging capacity. It is location. The Interstate Oil and Gas Compact Commission counts 141,959 documented orphaned wells on state and private land, up 54% from 92,203 in 2020, and estimates a further 250,000 to 740,000 undocumented orphaned wells whose positions are unknown. Lawrence Berkeley National Laboratory independently estimates 310,000 to 800,000 (Ciulla et al., 2024). A well that cannot be located cannot be inspected, contracted, or plugged, and the undocumented inventory runs 1.8 to 5.2 times the documented one.
The environmental and public-health burden accrues continuously while these wells remain unlocated. DOI estimates the combined population emits roughly 63 million grams of methane per hour, equivalent to the annual emissions of more than 3.6 million gasoline-powered passenger cars. Direct field measurement at Oil Creek State Park in Pennsylvania found that 62% of ground-verified wells remained unplugged and accounted for the majority of leaking wells, at a mean emission factor of 0.027 ± 0.099 kg/day (Saint-Vincent et al., 2021). Unplugged wells also act as vertical conduits between hydrocarbon reservoirs and shallow groundwater, so the unmet need is simultaneously a climate-forcing problem and a drinking-water contamination problem for the rural communities sitting above these fields.
2. State of the Art
Three research communities work this problem, and the boundary between what each has achieved and what each has left open defines the opportunity.
Records and imagery mining
The Earth and Environmental Sciences Area at Lawrence Berkeley National Laboratory, under Charuleka Varadharajan with Fabio Ciulla as lead author, treats candidate generation as a computer vision problem over historical cartography: a U-Net detects the standardized symbols denoting wells on georeferenced USGS historical topographic maps, and any symbol more than 100 m from a well in a modern regulatory database is flagged as a candidate. The method is nationally scalable because the map series is national. Los Alamos National Laboratory, under Hari Viswanathan and Daniel O’Malley, works the parallel imagery channel and has invested in shared infrastructure: a released corpus of 120,948 aerial images at 512 × 512 with segmentation masks, from National Agriculture Imagery Program coverage at 30 cm to 1 m ground sample distance (Kim et al., 2024).
Aerial magnetometry
Alex Nikulin and Timothy de Smet at Binghamton University established the operating envelope all subsequent work assumes, flying a miniaturized total-field magnetometer over historical fields in New York State. Patricia Saint-Vincent, James Sams, and Richard Hammack at the National Energy Technology Laboratory scaled the method and quantified how badly the regulatory record understates reality: across six aeromagnetic surveys in Pennsylvania and Wyoming, every site returned more magnetic points than the state database recorded. The same group demonstrated multi-modal survey at Oil Creek State Park, pairing airborne magnetics with LiDAR in terrain too steep and vegetated for ground search. The CATALOG consortium settled the platform question: fixed-wing drones carrying magnetometers are the most cost-effective configuration for discovering unknown wells, ahead of both helicopters and quadcopters (O’Malley et al., 2024).
Methane sensing and plume localization
W. H. Thomas and Caixia Wang at the University of Alaska Anchorage attack the wells magnetometry cannot see, on the principle that a leaking well announces itself regardless of casing material; their contribution is statistical anomaly detection robust to the noise and drift that make field gas-sensor data difficult. Mostafa Hassanalian and Sihua Shao at New Mexico Institute of Mining and Technology, with Xiang Sun at the University of New Mexico and Derek Hollenbeck, work the step beyond detection: given a plume, where is the source? Their multi-agent reinforcement learning framework coordinates several UAVs through virtual anchor nodes and infers the source from the trajectory of those placements. It is the most directly relevant embodied-AI result in this literature, and it exists only in simulation.
3. Foundational Research
Marcato A, Colman R, Milazzo D, Guiltinan EJ, Ma Z, O’Malley D, Viswanathan H, Santos JE. (2026). “Synthetic Training Enables Deployment on Raw Drone Data: An Attention-Based Framework for Detecting Orphan Wells.” Sensors, 26(9), 2573. DOI: 10.3390/s26092573.
Method and results. A transformer detector for magnetometer traces that avoids the heavy downsampling conventional pipelines require, using on-the-fly windowing and sinusoidal positional encoders for relative positional awareness along the flight line. Training is entirely on synthetic magnetic-anomaly data, with real drone data reserved for evaluation. On synthetic data the model sustains an F1 score above 90% at flight-line spacings up to 140 m, an operationally load-bearing figure because line spacing is the dominant cost driver in aerial survey. On real-world drone data the same model achieves 70% recall. The authors record that existing processing methods require heavy preprocessing and achieve unsatisfactory recall.
Ciulla F, Santos A, Jordan P, Kneafsey T, Biraud S, Varadharajan C. (2024). “A Deep Learning Based Framework to Identify Undocumented Orphaned Oil and Gas Wells from Historical Maps: A Case Study for California and Oklahoma.” Environmental Science & Technology, 58(50), 22194–22203. DOI: 10.1021/acs.est.4c04413.
Method and results. A U-Net detects well symbols in georeferenced historical topographic maps; any symbol more than 100 m from a well in a modern regulatory database is classified as a candidate. Across four counties in California and Oklahoma spanning more than 40,000 km², the framework identified 1,301 potential undocumented orphaned wells, of which 29 were confirmed from satellite imagery and 15 by field magnetic survey, at spatial accuracy on the order of 10 m. The paper supplies the working national estimate of 310,000 to 800,000 undocumented orphaned wells.
Saint-Vincent PMB, Sams JI III, Hammack RW, et al. (2020). “Identifying Abandoned Well Sites Using Database Records and Aeromagnetic Surveys.” Environmental Science & Technology, 54(13). DOI: 10.1021/acs.est.0c00044.
Method and results. Six aeromagnetic surveys in Pennsylvania and Wyoming, with detected magnetic points compared against state regulatory databases and the systematic divergence used to build a statistical estimator of true well counts. At every one of the six sites, more magnetic points were detected than the databases recorded. Estimates: 395,000 to 466,000 total wells in Pennsylvania, 181,000 to 182,000 in Wyoming, and a continental United States average of 6.04 ± 19.97 million wells, of which 1.16 ± 3.84 million are abandoned.
Saint-Vincent PMB, Sams JI III, et al. (2021). “Historic and modern approaches for discovery of abandoned wells for methane emissions mitigation in Oil Creek State Park, Pennsylvania.” Journal of Environmental Management, 280, 111856. DOI: 10.1016/j.jenvman.2020.111856.
Method and results. Airborne magnetic and LiDAR surveys over the Pioneer Run watershed, terrain too steep and vegetated for systematic ground search, corroborated against historical maps and georeferenced nineteenth-century photographs, then ground-verified and monitored for methane. LiDAR identified 290 field locations, of which 86% proved to be possible well sites; 62% of ground-verified wells remained unplugged and comprised the majority of leaking wells; the mean emission factor for unplugged wells was 0.027 ± 0.099 kg/day. In the oldest fields, casing had been removed or wood conductor casing installed.
Thomas WH, Wang C. (2026). “Drone-Based Statistical Detection of Methane Anomalies Around Abandoned Oil and Gas Well Sites.” Sensors, 26(7), 2205. DOI: 10.3390/s26072205.
Method and results. A drone carrying a high-sensitivity methane sensor, with statistical procedures built for the noise and drift of field environmental sensor data, validated in two stages: controlled release experiments with known emission points, then a field site containing three abandoned wells with sparse emission profiles. The method reliably detects statistical methane anomalies under field conditions and reduces the scope of subsequent ground survey. It supplies a national denominator of approximately 3.9 million documented wells whose recorded locations and plugging status vary widely in accuracy, and explicitly names source attribution and localization as development still required.
Li Z, Hollenbeck D, Wu R, Sherman M, Shao S, Sun X, Hassanalian M. (2026). “Multi-Agent Reinforcement Learning for UAV-Based Chemical Plume Source Localization.” arXiv:2603.11582 [eess.SY], 12 March 2026.
Method and results. A multi-agent deep reinforcement learning framework for chemical plume source localization, motivated explicitly by undocumented orphaned wells and by the stated limitation that magnetometry often fails to detect older wells effectively. UAV navigation is coordinated through virtual anchor nodes, each aircraft sensing gas concentration and wind velocity both onboard and through shared measurements, with source identification analyzing the historical trajectory of anchor placements within the plume. Benchmarked against fluxotaxis, the established plume-tracing heuristic, it achieves superior localization accuracy and operational efficiency. All evaluation is in simulation.
4. Competitive Landscape
Commercial activity sits adjacent to the research problem rather than on top of it. Zefiro Methane Corp. (Cboe Canada: ZEFI) plugs orphaned wells and monetizes avoided methane as carbon credits, and assembled its locating capability through partnership rather than internal development: a January 2025 letter of intent with Geolabe for AI processing of satellite methane imagery, a December 2024 engagement with Keynum for an inventory dashboard, and a provisional patent filed by sponsor entity X Machina Sustainable Technologies for an orphaned well data portal. The approach is satellite-down and records-based, and satellite methane instruments resolve at scales far coarser than a wellbore, which is why a field verification step survives in that workflow exactly as it does in the research pipelines. SPH Engineering supplies UgCS ground-station software with terrain-following flight planning and magnetometer payload support, and Geometrics (MagArrow), Sensys, and GEM Systems supply the instruments, including the Sensys R3 flown in published surveys at roughly 40 m. These are pre-flight planning tools and sensors; none performs in-flight inference.
From a research-funding standpoint the relevant competition is not commercial. It is the CATALOG consortium, and the resource asymmetry should be stated without euphemism: five national laboratories operating a $30 million program have produced the best detection models, the best public datasets, and the platform guidance the field follows, and no small team will out-resource them on detection science. The argument for a separate program is that CATALOG’s charter does not extend to flight-integrated deployment or manufacturable hardware, so the translational step proposed here falls outside what the consortium is constituted to deliver. The productive posture is complementary: this program consumes CATALOG’s published models and datasets as inputs and returns a deployment pathway for them.
5. Total Addressable Market
The market is autonomous aerial survey services and onboard autonomy systems for locating and positionally verifying undocumented orphaned oil and gas wells in the United States and Canada. Framed as public benefit at scale, which is how a federal funder should read it: locating even the low end of the undocumented inventory would place 250,000 wells within reach of an appropriation that already exists and currently cannot spend against them.
Bottom-up. Every input is drawn from primary federal reporting or peer-reviewed literature, and every derived figure is labeled as derived. The appropriated pool is $4.677 billion under IIJA Section 40601, of which approximately $1.8 billion was obligated through 30 September 2025, leaving roughly $2.9 billion unobligated. The State Program, holding $4.275 billion, explicitly covers identifying and characterizing undocumented orphaned wells alongside plugging them. DOI does not break out the locating share, and two independent anchors bound it. First, the Department of Energy has separately committed $45 million to locating and characterization alone: $30 million to CATALOG under IIJA, plus up to $15 million announced 13 September 2024 for Undocumented Orphaned Well Characterization and Remediation, structured as up to seven awards of $1.2 million to $3.4 million with 20% minimum cost share. Second, applying a 3% to 8% locate-and-characterize share to the $4.1 billion in direct State Program funding, a band consistent with site-characterization fractions in comparable environmental remediation programs and stated here as an assumption rather than a sourced figure, yields $123 million to $328 million across the program's remaining life, or roughly $20 million to $55 million per year through 2031.
Top-down cross-check. The methane emissions monitoring market was valued at $4.8 billion in 2025 and is projected to reach $11.6 billion by 2034, a 10.3% CAGR, with the software segment at approximately 31.5% of 2025 revenue and growing fastest at 13.7% CAGR (Dataintelo, Methane Emissions Monitoring Market Research Report, 2025). Orphaned well location is a small and underserved segment inside that envelope.
SAM. At TRL 4, with no certified beyond-visual-line-of-sight mission flown, the serviceable slice over 36 months is the $45 million in DOE technology-development funding plus early survey contracts in the three to five states holding both large undocumented inventories and active Formula Grant Phase 2 draws. At a 10% to 15% share of the midpoint services band, that is approximately $3 million to $8 million in annual services revenue by year four.
Payment pathway. No CPT or HCPCS reimbursement analogue applies, because this is not a clinical service; the equivalent question is which mechanism pays for deployment after the grant ends, and there are three documented answers. Grants flow to states through the DOI Orphaned Wells Program Office as Initial Grants ($570 million across 26 states), Formula Grants (more than $520 million in Phase 1, up to $826 million in Phase 2), and Matching Grants ($78 million across five states). Direct federal contracting flows through the bureaus: BLM $35 million, BSEE $46 million, FWS $64 million, USFS $58 million, NPS $39 million. The third rail is the voluntary carbon market via the American Carbon Registry methodology for plugging orphaned oil and gas wells, published May 2023 as the first of its kind covering the United States and Canada, and it carries a caveat any funder will want stated: ACR made version 1.0 inactive effective 9 May 2025 pending revisions, with already-verified projects and previously issued credits unaffected. The commercialization case here does not depend on it.
6. Research Gap and HHA Contribution
What has not been done
Every locating workflow currently deployed is open loop. The aircraft executes a mission file composed before takeoff, records data, lands, and surrenders that data to a pipeline whose output arrives days later. It makes no decision, tests no hypothesis, and adapts to nothing it observes. This is the natural product of a division of labor in which geophysicists own the sensor, machine learning researchers own the detector, and neither owns the aircraft.
The consequences are quantified in the literature itself. Detection recall on real drone data is 70% against an F1 above 90% on synthetic data, and because inference is offline the missed 30% are never revisited (Marcato et al., 2026). Candidate confirmation runs at roughly 3%, 44 wells from 1,301 candidates (Ciulla et al., 2024). And the two channels that would cover each other's blind spots are never fused during flight, so a well whose casing was pulled and which is actively venting is simultaneously invisible to the magnetic detector and unlocalizable by the methane sensor (Saint-Vincent et al., 2021; Thomas and Wang, 2026).
The precise technical gap
No system performs onboard multimodal inference and in-flight re-tasking during an orphaned well survey. Three capabilities are missing, each for a different reason. Onboard magnetic inference does not exist because the published transformer was designed for offline throughput on recorded traces, not an embedded accelerator inside a fixed-wing payload's weight and power budget. Cross-modal arbitration does not exist because the magnetic and methane communities publish separately and no shared decision layer has been specified. Flight-adapted plume localization does not exist because the policy that solves it was evaluated entirely in simulation, with no wind-estimation error model, sensor lag model, or endurance constraint.
The proposed research program
The program runs four workstreams. Workstream 1 produces an embedded-deployable variant of the published transformer through quantization and knowledge distillation, targeting sustained inference at the magnetometer's native sample rate within a defined power envelope, with recall retention within 5 percentage points of the offline model as the controlling metric. Workstream 2 implements the arbitration logic letting either channel initiate a confirmation maneuver, with methane as an independent trigger rather than a channel subordinate to magnetics, since a pulled-casing well emits gas but no magnetic anomaly. Workstream 3 ports the multi-agent reinforcement learning policy onto a real airframe with domain randomization over wind-estimation error, sensor lag, and endurance limits the simulation did not model, benchmarked against fluxotaxis using controlled releases at surveyed coordinates. Workstream 4 designs the integrated magnetometer, methane sensor, and compute payload for repeatable production from the first prototype. Evaluation throughout is a two-arm field trial over instrumented targets of both classes against an open-loop grid survey of the identical targets, and the headline metric is cost per confirmed well.
Why HHA is positioned to do this
Each gap maps to a specific team member. Onboard inference and localization-policy adaptation map to Haedar Hadi, whose MS in Computer Science from Boston University with an Information Systems focus and background in scalable compute architecture cover both the model-compression work and the evaluation methodology that makes a recall-retention claim defensible. Sensing physics, cross-modal arbitration, and controlled-release experimental design map to Hass Dhia, whose training spans the physical sciences and whose work in sensor fusion and autonomous navigation addresses the magnetic anomaly interpretation, plume physics, and wind-field estimation the decision layer consumes. The manufacturable payload maps to Ahmed, Director of Manufacturing, and it is the workstream academic proposals in this class routinely omit.
Why the originating laboratories have not closed this gap
Mandate, not capability and not funding. CATALOG's charter is to develop technologies and methodologies enabling the finding and characterizing of undocumented orphaned wells, and the consortium discharges it exactly as constituted: publishing models, releasing datasets such as the 120,948-image corpus, and issuing platform guidance. A national laboratory has no product function, no design-for-manufacturability discipline, and no route to airworthiness certification, because none of those is what a national laboratory is instrumented to produce. The published artifacts show the boundary: the Los Alamos detector was evaluated offline on recorded traces, and nothing in that paper concerns latency budgets, power envelopes, or embedded targets, because those questions belong to a deployment program that does not exist.
7. Comparable Funded Projects
CATALOG: Consortium Advancing Technology for Assessment of Lost Oil & Gas Wells. Funder: DOE Office of Fossil Energy and Carbon Management, under IIJA. $30 million. Institutions: Los Alamos, Lawrence Berkeley, NETL, Sandia, and Lawrence Livermore national laboratories. Kickoff April 2022; five-year program. Every national-laboratory result cited in Section 3 traces to it, which establishes that DOE treats locating undocumented wells as a fundable research problem in its own right rather than an administrative line inside plugging contracts. It is best positioned as an input and collaboration target rather than a competitor.
Undocumented Orphaned Well Characterization and Remediation. Funder: DOE Office of Fossil Energy and Carbon Management. Up to $15 million, announced 13 September 2024, structured as up to seven awards of $1.2 million to $3.4 million each with 20% minimum cost share; applications due 13 November 2024. Focus areas: advanced borehole remediation, wellbore characterization, and long-term monitoring. This is the closest match to the instrument the proposed work would seek, and eligibility explicitly extends to for-profit organizations, universities, non-profits, and state, local, and tribal governments, so an independent research institute is a directly eligible applicant.
NSF Award #2323050, “SCC-PG: Sustainable Vertiports for Bringing Autonomous Drone Swarm Inspection to Oil and Gas Industry.” PI: Sihua Shao, New Mexico Institute of Mining and Technology. $149,999, 1 October 2023 to 30 September 2025, NSF Smart and Connected Communities Planning Grant. The most directly relevant funded analogue, and the connection is not coincidental: PI Shao co-authored the 2026 multi-agent reinforcement learning plume-localization paper cited in Section 3, so the same NSF-funded group that produced the control policy is also working autonomous drone infrastructure for oil and gas.
NSF Award #2133029, “NRI: An Ecologically Curious Robot for Monitoring Coral Reef Biodiversity.” PI: Yogesh Girdhar, Woods Hole Oceanographic Institution. $1,499,986, 1 March 2022 to 31 December 2025, NSF National Robotics Initiative. The methodological rather than domain neighbor and the most useful comparable for sizing a request: an autonomous vehicle deciding in the field what to observe next from what it has already observed, which is informative path planning under a sensing budget in an unstructured environment. It establishes both the program and the scale, roughly $1.5 million over four years.
NSF Award #2128403, “I-Corps: Autonomous electronic sentinel microsensors for monitoring and logging harsh environments.” PI: David Chapman, University of Texas at Austin. $50,000, 1 April 2021 to 31 March 2023. Small in dollars but significant as a signal, since I-Corps funds customer discovery rather than research: NSF has already paid to validate that a commercial customer exists in this segment.
One sourcing limitation is recorded rather than concealed: the SBIR.gov awards API returned HTTP 403 Forbidden on every query attempted during this analysis, so SBIR and STTR awards are not represented above. Their absence reflects a data-access failure, not an absence of relevant awards, and that landscape should be surveyed before submission.
8. Opportunity Assessment
TRL assessment
The integrated system is at TRL 4, with components spanning TRL 3 to 6. Aerial magnetic detection of well casing is field-validated at TRL 5 to 6: anomalies confirmed from the surface to 50 m AGL with 40 m optimal, across six NETL surveys in two states and independent UAV work in New York. Machine learning detection on magnetometer data is at TRL 4, validated on real drone traces at 70% recall but executed offline. Methane anomaly detection is at TRL 4 to 5, validated against controlled releases and tested at a three-well field site. Autonomous plume source localization is at TRL 3, benchmarked in simulation only. The integrated closed-loop system exists at no readiness level, which places the system as a whole at TRL 4; advancing to TRL 5 requires a flight-integrated payload demonstrating in-flight re-tasking in a relevant outdoor environment against instrumented targets.
Technical risks framed as research questions
Can a transformer detector trained on synthetic magnetometer data run onboard at the sensor's native sample rate without losing the recall that justifies flying it? Quantization and distillation to an embedded accelerator is the proposed answer, with a go/no-go at month 6: recall within 5 percentage points of the offline baseline on held-out real drone traces, or pivot to lightweight anomaly triggering with deferred inference.
Does in-flight re-tasking improve detection yield, or merely redistribute survey time? A confirmation maneuver costs endurance that would otherwise cover new ground, so the benefit must be measured rather than assumed. The experiment is a paired two-arm field trial over the same instrumented targets, with confirmed wells per flight hour as the primary endpoint and cost per confirmed well as the secondary.
Can multi-agent plume localization survive the transition from simulation to a real wind field? The policy consumes wind velocity estimates that are clean in simulation and noisy in flight. Domain randomization over wind-estimation error, sensor lag, and endurance constraints is the mitigation, with the fluxotaxis benchmark re-run against controlled releases at surveyed coordinates.
Regulatory pathway
This system is not FDA-regulated and no 510(k), De Novo, or PMA pathway applies; the governing authority is the Federal Aviation Administration. Research flights operate under 14 CFR Part 107, which restricts operations to visual line of sight absent a case-by-case waiver. The consequential change is proposed Part 108: the FAA published the NPRM “Normalizing Unmanned Aircraft Systems Beyond Visual Line of Sight Operations” on 7 August 2025, the comment period closed 6 October 2025 with more than 3,000 comments, and the FAA reopened comment on 28 January 2026 on ADS-B Out equipage, alternate electronic conspicuity devices, and detect-and-avoid requirements. Final rule publication is expected during 2026, implementation 6 to 12 months afterward. Surveys over federal lands require bureau land-management authorization, and ground crews approaching potential methane emitters fall under OSHA hazardous-atmosphere provisions. A 24-month program is well-timed: the research is executable under Part 107 throughout, and the system reaches demonstration maturity as the BVLOS framework becomes operative. The burden is a moat as much as a cost, since Part 108 will impose airworthiness, operator certification, and detect-and-avoid obligations a component vendor or small survey shop cannot readily satisfy.
Locked versus adaptive algorithm
This determines the certification path and must be settled at design time rather than discovered at review. A model that continues learning on-aircraft is, from a certifying authority's standpoint, a system whose behavior at flight hour 1,000 is not the behavior certified at hour zero. The proposed architecture is a locked inference model in the flight-critical path, retrained and revalidated on the ground under version control, with adaptation confined to the non-flight-critical survey-planning layer within pre-declared bounds. This mirrors the FDA Predetermined Change Control Plan framework for adaptive medical AI, under which the permissible envelope of post-deployment modification is agreed in advance. Aviation has no equivalent framework yet, and voluntarily adopting the PCCP discipline is both the defensible engineering choice and a differentiator before a reviewer who knows the medical-device precedent.
9. Team Fit
Hass Dhia, Co-Principal Investigator
MS Biomedical Sciences with a medical school background including service as an anatomy teaching assistant, an operating history as an AI infrastructure architect, and a working span across the physical sciences: physics, chemistry, fluid dynamics, environmental and energy systems, and UAV system design across scales. The mapping is to sensing and experimental design. Magnetic anomaly interpretation against the 40 m altitude envelope, methane plume physics and the wind-field estimation the localization policy consumes, and the cross-modal arbitration logic are physical-systems problems before they are software problems, and the controlled-release validation trials that anchor every performance claim are experimental-design problems. He also owns the problem-framing interface with state regulatory agencies, whose operational definition of a located well is what the program's endpoint must match.
Haedar Hadi, Lead Principal Investigator
MS Computer Science, Boston University, with an Information Systems focus, and a professional background in cloud and database architecture. The mapping is to Workstreams 1 and 3. Model compression, quantization, and distillation of the published transformer to an embedded inference target is ML engineering, and the scalable-compute background makes the training and domain-randomization pipeline tractable at the data volumes magnetometer surveys generate. Equally important and more often overlooked, he owns evaluation methodology and benchmark design: the program's central claims are comparative, and a comparative claim is only as credible as the benchmark that produced it.
Ahmed, Director of Manufacturing
Director of Manufacturing, with expertise in design for manufacturability, production scaling, quality systems, process optimization, and industrial automation. The mapping is to Workstream 4 and, through tolerance and producibility constraints, to payload decisions inside every other workstream.
The lab-to-production bridge. Most research proposals end at “it works in the lab.” This proposal includes explicit DFM milestones at every phase, ensuring that prototype decisions consider production scaling, tolerance analysis, and quality systems from day one. This addresses the valley of death between TRL 4-5 prototypes and TRL 7+ deployable systems, the gap where most funded research stalls. Here specifically, that discipline determines whether the sensor payload can be replicated across the fleet a state survey contract would require, and whether the build documentation will satisfy an FAA airworthiness review rather than having to be reconstructed after the fact.
Capability gaps and how funding addresses them
The team holds no airframe manufacturing capability, no Part 107 commercial survey operations history, and no in-house geophysical crew. The budget in Section 10 addresses each explicitly: subcontracted flight operations with an established Part 107 survey operator, a subaward to a university geophysics group for independent ground-truth verification, and commercial off-the-shelf airframe and magnetometer procurement rather than in-house development. The team is positioned as the autonomy and manufacturability layer atop a mature instrument and airframe supply chain, not as a competitor to the national laboratory consortium on detection science or to survey contractors on flight operations.
10. Recommended Next Steps
Target funder programs
DOE Office of Fossil Energy and Carbon Management, Undocumented Orphaned Well Characterization and Remediation (successor solicitation to the FOA announced 13 September 2024, administered through NETL). The primary target: the $1.2 million to $3.4 million award structure with 20% cost share matches the scope proposed here, for-profit and non-profit organizations are explicitly eligible, and the monitoring and characterization focus areas map directly onto the survey and verification work.
NSF National Robotics Initiative / Foundational Research in Robotics. The natural home for Workstreams 1 and 3 framed as fundamental research in informative path planning and sim-to-real transfer for field robotics, with NSF Award #2133029 establishing precedent and scale.
NSF SBIR/STTR Phase I, Environmental and Biological Sciences topic. The commercialization track for the payload and autonomy stack, with I-Corps award #2128403 in this exact application area as evidence of validated customer demand.
Estimated funding range
Based on the comparable awards in Section 7, the defensible ask is $1.5 million to $2.5 million over 24 months. The DOE FOA band of $1.2 million to $3.4 million per award brackets it directly, and the NRI comparable of $1.5 million over four years supports the implied annual burn. A request materially above $3 million would exceed every observed comparable; a request below $1 million would not fund the flight operations subcontract the two-arm field trial requires.
Proposed 24-month milestone timeline
R&D track, months 1 to 18. M1–M3: acquire CATALOG datasets, independently reproduce the published 70% real-data recall baseline, instrument a test range with surveyed ferrous-cased and non-ferrous targets. M4–M6: first distilled onboard model, bench-validated at native sample rate within the target power envelope. Go/no-go at M6: recall within 5 percentage points of offline baseline, or pivot to lightweight anomaly triggering with deferred inference. M7–M12: cross-modal arbitration implemented; reinforcement learning policy ported with domain randomization. M13–M18: integrated flight testing under Part 107, then the two-arm field trial over identical instrumented targets. Go/no-go at M18: confirmed wells per flight hour must exceed the open-loop arm by a pre-registered margin.
Regulatory track, months 1 to 24, parallel. Part 107 operations established through the flight subcontractor by M6, with airworthiness documentation practices defined alongside the manufacturing lead so the build record is generated as the build happens. Locked-model version control and validation protocol formalized against the PCCP-analogous framework by M18. Part 108 readiness assessment and operator certification package prepared M19–M24 against the final rule as published.
Manufacturing and DFM track, months 1 to 24, beginning at month 1. Payload requirements and producibility constraints defined before the first prototype is cut, including tolerance stack-up analysis for the sensor mounting geometry that sets the magnetic noise floor (M1–M3). Preliminary design review and supplier qualification for magnetometer, gas sensor, and compute module (M4–M9). Critical design review, production build documentation, and a fleet-replication cost model at 5, 20, and 50 units (M10–M18). Pilot production run of three units for the field trial fleet, with first-article inspection protocol and a quality plan sized to a state survey contract (M19–M24).
Path to first revenue. The two-arm field trial completing at M18 is the artifact supporting a state survey contract bid, and the pilot production run makes such a bid deliverable. First revenue is targeted in the quarter following program completion, contingent on the M18 go/no-go, and does not depend on Part 108 having taken effect.