1. Safety Need
Underground blasting produces toxic gases: carbon monoxide (IDLH 1,200 ppm), nitrogen dioxide (IDLH 20 ppm), methane (explosive at 5 to 15% concentration), and sulfur dioxide (IDLH 100 ppm). After each detonation, these gases disperse through mine workings at rates governed by ventilation geometry, blast energy, and geological conditions. Current practice enforces conservative re-entry waiting periods of two to four hours, then sends human inspection teams with handheld gas monitors to verify air quality.
MSHA recorded 10 miner fatalities between January 3 and March 5, 2025, more than triple the rate from the same period in 2024. Globally, underground mining fatalities from toxic gas exposure number in the hundreds annually. China reported 4,746 coal mining deaths in 2006, with most victims dying from gas inhalation.
Over 2,300 active underground mines operate worldwide (GlobalData, 2024). Each conducts multiple blast cycles per shift. Cumulative re-entry delays reduce productive mining time by 15 to 30%, costing $50,000 to $150,000 per blast event in deferred production.
The research need is an autonomous system that enters blast zones within minutes of detonation, maps three-dimensional gas concentration fields, and provides quantitative re-entry safety assessment based on measured data rather than conservative time estimates.
2. State of the Art
Autonomous post-blast gas measurement
Nordstrom et al. (2025, Journal of Field Robotics, DOI: 10.1002/rob.22500) at Lulea University of Technology, Sweden, developed the RIA framework: risk-aware 3D path planning combined with LiDAR-based global relocalization and onboard gas sensors. This represents the first autonomous post-blast gas measurement. The system operated at the K+S Group potash mine in Germany at 700m depth, 40 minutes post-blast, successfully navigating through dust and structural deformation.
Reconfigurable borehole-deployable drones
Brown et al. (2026, Journal of Field Robotics, DOI: 10.1002/rob.70112) at the University of Manchester developed Prometheus, a reconfigurable drone that folds to 130mm diameter for borehole deployment and unfolds to 815mm for autonomous flight. Validated at Holman’s test mine in Cornwall, UK, the system achieved beyond visual line of sight operation at depths exceeding 19m.
Multi-robot collaborative mine inspection
Lulea University and LKAB (2024) demonstrated the first multi-robot autonomous mine inspection using three drones collaborating at the Konsuln mine in Sweden. The system operated in GPS-denied conditions using LiDAR SLAM navigation, published via Lulea University press release and ArcticToday.
DARPA Subterranean Challenge
The DARPA Subterranean Challenge (2021) brought together 14 teams for multi-robot underground exploration. Team CERBERUS from the University of Nevada Reno won the $2M prize, validating autonomous underground navigation with commercial hardware.
Quadruped navigation in mines
A 2025 study (arXiv:2603.04470) demonstrated Boston Dynamics Spot operating at the Lulea/LKAB experimental mine, achieving 700+ meters of autonomous traverse with 100% success rate across 20 trials on edge hardware without GPU.
Adjacent commercial systems
Exyn Technologies ($35M+ raised, Philadelphia) and Emesent (Brisbane) offer commercial mine mapping and survey solutions. Both focus exclusively on mapping. Neither provides gas detection capabilities.
The gap: no existing system combines autonomous GPS-denied navigation with multi-gas sensing for post-blast re-entry assessment.
3. Foundational Research
Nordstrom et al. (2025). Journal of Field Robotics. DOI: 10.1002/rob.22500.
RIA framework for risk-aware autonomous post-blast inspection. First demonstrated autonomous gas measurement in a production mine environment at K+S potash mine, 700m depth, 40 minutes after blast detonation. The system measured CO and NO2 concentrations in post-blast conditions characterized by heavy dust and map deformation from blast damage. LiDAR-based global relocalization enabled navigation when pre-blast maps no longer matched the physical environment.
Brown et al. (2026). Journal of Field Robotics. DOI: 10.1002/rob.70112.
Prometheus reconfigurable drone platform. Folds to 130mm diameter for deployment through standard mining boreholes, unfolds to 815mm wingspan for autonomous flight. Validated at Holman’s test mine in Cornwall, UK, at depths exceeding 19m with beyond visual line of sight operation. Solves the critical access problem for mine voids that are physically inaccessible to human inspectors and conventional drones.
Lulea University / LKAB (2024). Multi-robot autonomous mine inspection.
First demonstration of collaborative multi-robot autonomous inspection in an active underground mine. Three drones operated simultaneously at the Konsuln mine in Sweden under full GPS-denied conditions. LiDAR SLAM provided cooperative localization. The multi-robot architecture enables parallel coverage of large blast zones, reducing inspection time proportionally to fleet size.
Sganderla et al. (2025). PMC. Comprehensive survey of autonomous mine inspection robots.
Systematic review identifying the primary barriers to deployment of autonomous inspection systems in underground mines. Key challenges include dust interference with optical sensors, electromagnetic interference from geological formations and mining equipment, battery endurance limitations in low-oxygen and high-temperature environments, and the absence of regulatory frameworks governing autonomous robot operation in occupied mines.
DARPA Subterranean Challenge (2021). Team CERBERUS, University of Nevada Reno.
Multi-year competition culminating in the Final Event where Team CERBERUS located 23 of 40 artifacts across tunnel, urban, and cave environments using a multi-robot autonomous system. Validated that commercial-grade hardware (LiDAR, IMU, compute modules) can support reliable autonomous underground navigation at mission-relevant scales. The competition produced open-source software stacks now available to the research community.
4. Competitive Landscape
Two companies operate in the adjacent space of autonomous mine mapping. Exyn Technologies ($35M+ raised, Philadelphia) provides autonomous drone mapping for underground environments. Emesent (Brisbane) offers Hovermap, a LiDAR-based autonomous mapping payload. Both companies focus exclusively on geometric mapping and survey. Neither offers gas detection, post-blast assessment, or re-entry safety scoring.
No commercial entity currently offers an integrated system combining autonomous GPS-denied navigation with multi-gas sensing for post-blast re-entry assessment. The reason is timing: the convergence of capabilities required (autonomous underground flight, onboard gas sensing, risk-aware path planning) was only demonstrated in 2025 by academic groups. The commercial gap reflects the recency of the underlying research, not a lack of market demand.
5. Addressable Scope
Bottom-up calculation
- Active underground mines worldwide: 2,300 (GlobalData, 2024)
- Average blast events per mine per day: 2
- Operating days per year: 300
- Total annual blast events: 1.38 million
- System unit cost: $200K–$400K per mine + $50K annual service
- Hardware TAM: ~$690M. Recurring service TAM: ~$115M/year.
Top-down cross-check
The global drone inspection market was valued at $9.1 billion in 2021 and is projected to reach $33.6 billion by 2030 at 15.7% CAGR (MarketsandMarkets). The mining drones segment was valued at $1.96 billion in 2023, projected to reach $10.67 billion by 2030 (Valuates). Post-blast autonomous inspection represents an estimated 5–10% of the mining drone market, yielding $500M–$1.1B by 2030. This is consistent with the bottom-up estimate.
Serviceable addressable market
Tier 1 mining operators (~500 mines with the capital and safety culture for early adoption) at an average system price of $300K yield an initial SAM of $150M.
6. Research Gaps and HHA Contribution
Three specific gaps separate published laboratory and field results from a deployable post-blast safety system:
Gap 1: Risk-aware adaptive navigation in degraded conditions
Post-blast environments degrade pre-existing mine maps. Rock displacement, dust loading, and structural deformation mean that a drone navigating on a pre-blast 3D map encounters geometry that no longer matches. Nordstrom et al. (2025) addressed this with LiDAR relocalization, but their system uses fixed path planning rather than adaptive policies.
Research question: Can a reinforcement learning agent navigate using degraded LiDAR and IMU inputs, combined with gas gradient signals, when the pre-blast map is inaccurate?
HHA approach: PPO-based adaptive navigation trained in simulated post-blast environments. Digital twin mine geometries generated from real mine LiDAR scans, with CFD gas dispersion models providing realistic gas concentration fields. The RL agent learns to use gas gradients as supplementary navigation signals when geometric features are obscured by dust or altered by blast damage. Haedar Hadi leads this work, drawing on ML and reinforcement learning expertise from the Boston University Computer Science program.
Gap 2: Multi-gas sensor fusion and quantitative re-entry assessment
Current systems measure individual gases at discrete points along a flight path. Mine safety decisions require volumetric gas maps with probabilistic safety scores across the entire blast zone, not point measurements from a single traverse.
Research question: Can an ML pipeline fuse multi-gas sensor data with ventilation models to predict gas clearance rates and compute probabilistic re-entry safety scores?
HHA approach: Bayesian sensor fusion with physics-informed machine learning. Gas transport equations (advection-diffusion models incorporating mine ventilation geometry) serve as inductive bias for the ML pipeline. Electrochemical sensor arrays measure CO, NO2, CH4, and SO2 simultaneously. The fusion pipeline produces volumetric gas concentration maps and time-to-safe-re-entry predictions with quantified uncertainty bounds. Hass Dhia leads this work. Biomedical sensing experience translates directly to electrochemical gas sensor interpretation, and physical sciences training in thermodynamics and fluid dynamics provides the foundation for ventilation modeling.
Gap 3: Ruggedized manufacturing for mine deployment
Academic prototypes that work in controlled test mines fail under production mining conditions. Dust infiltration, humidity, temperature extremes, vibration, and methane atmospheres destroy unprotected electronics and sensors within weeks.
Research question: What design-for-manufacturability choices enable mine-rated autonomous gas monitoring drones at fleet scale with ATEX compliance?
HHA approach: Ahmed leads manufacturing from day one as Director of Manufacturing. DFM analysis begins in parallel with algorithm development, not as an afterthought. Replaceable gas sensor cartridges with factory calibration enable field maintenance without specialized technicians. ATEX Zone 1 enclosure design addresses methane atmosphere requirements. Production tolerance analysis ensures consistent performance across manufacturing batches.
Why originating labs have not closed these gaps
Nordstrom’s group at Lulea is a robotics lab, not a sensor fusion or manufacturing group. Brown’s group at Manchester is mechanical engineering, not ML. The DARPA SubT teams have largely disbanded. No single academic lab has the combination of reinforcement learning algorithms, multi-sensor fusion, and manufacturing engineering required to close all three gaps simultaneously.
7. Comparable Funded Projects
| Source | PI / Entity | Amount | Focus |
|---|---|---|---|
| NIOSH RFA-OH-23-005 (2023) | Various PIs | $8M | Robotics and Intelligent Mining Technology |
| DARPA SubT Challenge (2018–2021) | 14 teams | Multi-year + $2M prize | Autonomous multi-robot underground exploration |
| NSF NRI 3.0 | Various PIs | $250K–$1.5M per project | Robotics integration, NIOSH co-funding for mine safety |
| Lulea / LKAB / SUM Academy | Lulea University | Multi-year | LKAB + Swedish Dept of Energy funded mine robotics |
| NIOSH/CDC FY2025 BAA | Various PIs | Variable | Applied Research for Mine Safety Technology |
8. Opportunity Assessment
TRL evidence chain
TRL 5, validated in relevant environment. Nordstrom et al. (2025) demonstrated autonomous post-blast gas measurement in a production potash mine at 700m depth. Brown et al. (2026) validated reconfigurable drone deployment through boreholes at a test mine. The DARPA SubT Challenge (2021) validated multi-robot autonomous underground navigation with commercial hardware across tunnel, urban, and cave environments. Collectively, these demonstrations establish that the individual subsystems (autonomous flight, gas sensing, underground navigation) function in mine-relevant conditions. The integration gap is combining all three in a single deployable system.
Top 3 technical risks
LiDAR SLAM accuracy in post-blast dust conditions
Mitigation: Nordstrom et al. (2025) already operated LiDAR SLAM through post-blast dust at the K+S mine. Their relocalization approach handled map deformation. The remaining question is performance in heavier dust loading at shorter post-blast intervals (under 20 minutes). Multi-modal sensing (IMU + gas gradient + degraded LiDAR) provides fallback when optical sensing is fully obscured.
ModerateRL agent generalization across different mine geometries
Mitigation: This is a transfer learning question. Training on digital twin models derived from multiple real mine LiDAR datasets provides geometric diversity. Domain randomization of gas dispersion parameters, dust density, and structural deformation builds robustness. Sim-to-real transfer validation at mine analog facilities quantifies the generalization gap before field deployment.
ModerateMulti-gas fusion precision for re-entry decisions
Mitigation: Electrochemical gas sensors are well-characterized instruments with known cross-sensitivity profiles. The sensor fusion challenge is spatial interpolation from sparse flight-path measurements to volumetric concentration maps. Physics-informed ML (using ventilation geometry as constraint) reduces the degrees of freedom. Controlled validation against fixed sensor arrays in mine analog facilities provides ground truth for fusion accuracy assessment.
ModerateRegulatory landscape
MSHA 30 CFR Part 18 governs electrical equipment in underground mines. ATEX Zone 1 and IECEx certification are required for operation in potentially explosive methane atmospheres. No existing regulatory framework addresses autonomous drones in occupied mines. Early engagement with MSHA (pre-submission meetings) represents a structural advantage: the operator that helps define the regulatory framework shapes the requirements that all subsequent entrants must meet. Certification timelines of 18 to 24 months create a durable moat for first movers.
First 6 months
(a) Build simulation environment using real mine LiDAR data with CFD gas dispersion models. (b) Train PPO/SAC navigation agents in simulated post-blast conditions. (c) Conduct DFM analysis for gas sensor cartridge design and ATEX-compliant enclosure.
9. Team Capabilities
Successful pursuit of this research direction requires three intersecting capabilities: adaptive autonomous navigation algorithms, multi-sensor fusion for safety assessment, and ruggedized manufacturing for mine deployment. HHA’s team provides coverage across all three:
Hass Dhia
MS Biomedical Sciences with medical school background (anatomy, physiology, pharmacology). AI infrastructure architect with production systems at scale. Maps to: gas sensor fusion algorithm design (biomedical sensing experience translates directly to electrochemical multi-gas interpretation), mine ventilation physics and gas transport modeling (physical sciences training in thermodynamics and fluid dynamics), experimental design for field validation, and regulatory strategy (MSHA pre-submission preparation). Trustworthy AI systems experience maps to safety-critical autonomous re-entry clearance decisions.
Haedar Hadi
MS Computer Science (Boston University, Information Systems focus). Specializes in ML model development, reinforcement learning architectures, and evaluation methodology. Maps to: adaptive navigation policy design for GPS-denied post-blast environments (PPO/SAC with degraded sensor inputs), sim-to-real transfer from digital twin mine models to physical deployment, multi-agent coordination for fleet-scale drone operations, and evaluation framework design for safety-critical autonomous systems.
Ahmed
Maps to: ruggedized drone airframe DFM for mine conditions (dust, humidity, temperature extremes, methane), ATEX/IECEx-compliant enclosure design, replaceable gas sensor cartridge manufacturing with factory calibration, and batch production quality systems.
Lab-to-production bridge
Most mine robotics 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, ATEX certification requirements, and quality systems from day one. This addresses the valley of death between TRL 4–5 prototypes and TRL 7+ deployable mine safety systems. The gap where most funded mine robotics research stalls is not the algorithm or the sensor. It is the transition from a hand-built research platform to a mine-rated product that operates reliably for thousands of hours in dust, water, and vibration. Ahmed’s manufacturing engineering background directly addresses this gap.
10. Recommended Next Steps
Target funding programs
| Program | Mechanism | Range | Fit |
|---|---|---|---|
| NIOSH Mining Safety | U60 Cooperative Agreement | $1M–$8M | Autonomous mine inspection, worker safety technology |
| NSF NRI 3.0 | Standard Grant | $250K–$1.5M | Robotics integration, human safety in mining |
| NIOSH/CDC BAA | Applied Research | Variable | Mine safety technology development and demonstration |
| NSF SBIR/STTR | Phase I / Phase II | $275K / $1M | Commercialization of mine safety autonomous systems |
| DARPA (SubT follow-on) | Research Agreement | $1M–$5M | Autonomous underground navigation and sensing |
Estimated total funding range: $1.5M to $5M over 24 months.
24-month milestone timeline
- M1–3 Simulation environment construction (mine LiDAR + CFD gas dispersion). DFM requirements analysis. Literature review and baseline benchmarks.
- M4–8 RL navigation agent v1 in simulation. Multi-gas sensor fusion pipeline. ATEX enclosure prototype. MSHA pre-submission meeting preparation.
- M9–14 Sim-to-real validation at mine analog facility. Gas sensor accuracy benchmarking. DFM prototype v1 (target: <10% rejection). MSHA meeting.
- M15–20 Field validation at partner mine (real post-blast). Manufacturing scale-up analysis (50+ units/batch). Quality system documentation.
- M21–24 Performance publication. Phase 2 funding application. ATEX certification submission.