Opportunity Assessments

Research briefs.

Where AI meets unmet need. Each brief evaluates a direction for scientific merit, technical readiness, team fit, and funding viability — across five scale tiers, from planetary systems to molecular machines.

Published Briefs

Research direction assessments

Each brief evaluates a research direction for scientific merit, technical readiness, team fit, and funding viability.

HECTO TRL 4 August 31, 2026

Autonomous Dual-Arm Robotics for Energized Electric Distribution Work

Contact with an overhead power line causes 49% of workplace electrical fatalities in the United States, and the occupation that does that work has an electrical-cause fatality rate of 6.01 per 100,000 against 0.11 across all occupations, while non-fatal electrical injuries rose 59% between the 2021–2022 and 2023–2024 reporting periods (ESFI, compiling BLS Census of Fatal Occupational Injuries data). The capability to remove the human is further along than the market suggests. An autonomous robot has connected and disconnected lead-flow lines on energized 10 kV overhead lines in field experiments at close to manual efficiency (Feng and Zhang, IEEE Access, 2021); a 12-degree-of-freedom dual-arm system has been brought under 150 kg so it rides the aerial platforms utilities already own (Wang, Industrial Robot, 2024); cooperative live-line strand repair ran 68.2% faster than manual work in laboratory tests (Tian et al., Electronics Letters, 2026); and the arc a robot draws entering equipotential state has been measured at 100 to 200 A with energy below 100 MHz (Pei et al., Scientific Reports, 2025). Every one of those results was produced against another country’s grid hardware and rules, every control architecture is a symbolic planner whose authors concede residual operator correction, and no United States framework exists for accepting a robot inside the OSHA minimum approach distance. Single-purpose placement robots already work energized lines commercially (FulcrumAir with PLP up to 240 kV; Meta’s Bombyx on energized medium-voltage distribution), which is the best evidence the general case is achievable and also the reason the market count is three rather than zero. What blocks the general case is not rate-base economics, a common but incorrect account, since a robot is itself capital: it is that cost-of-service returns an efficiency saving to ratepayers at the next rate case, so the business case cannot be denominated in wages, and that most energized work is bought from contractors whose revenue is billable crew hours.

Embodied AI

MEGA TRL 4 August 24, 2026

Adaptive Autonomous Underwater Sampling for Marine Carbon Dioxide Removal Verification

Buyers have contracted 578,000 tonnes of marine carbon dioxide removal and 0.3% of that volume has been issued as credits (AlliedOffsets, October 2025). The bottleneck is in-water verification. A WHOI-led tracer study followed an alkalinity patch that displaced 14.8 km in 36 hours while a research vessel sampled it once every 1,200 metres, and captured only 10% of the intended carbon uptake before the campaign ended (Subhas et al., Biogeosciences, 2025); its authors name the ability to track the intervention through space and time as the central challenge of in-water MRV. Both halves of a solution are separately validated. Lab-on-chip alkalinity sensors have flown on the Autosub Long-Range AUV in the Celtic Sea with residuals of 1–2 µmol/kg against 156 co-samples (Hammermeister et al., Environmental Science & Technology, 2025), and MBARI’s long-range AUV autonomously tracked a chlorophyll feature for four days without surfacing by locking onto a fast thermal proxy (Zhang et al., IEEE Journal of Oceanic Engineering, 2020). Nothing joins them. ARPA-E’s $36 million SEA-CO2 portfolio funded eleven projects, every one a sensor or a model, and none the decision layer that determines where the sensor should be.

Embodied AI

MEGA TRL 4 August 3, 2026

Closed-Loop Autonomous Aerial Survey for Locating Undocumented Orphaned Oil and Gas Wells

The Infrastructure Investment and Jobs Act appropriated $4.677 billion to plug orphaned wells, but 250,000 to 740,000 undocumented wells cannot be contracted for plugging because nobody knows where they are (DOI FY2025 Annual Report to Congress). Every deployed locating workflow is open loop: a drone flies a pre-planned grid, lands, and the data is processed days later. The best published detector falls from an F1 above 90% on synthetic magnetometer data to 70% recall on real drone data (Marcato et al., Sensors, 2026), and map-derived candidate generation produced 1,301 candidates of which 44 were confirmed (Ciulla et al., Environmental Science & Technology, 2024). Magnetometry cannot see wells whose casing was pulled or was wooden, and methane sensing detects plumes it cannot localize. A multi-agent reinforcement learning policy that solves the localization step beats fluxotaxis in simulation and has never flown. DOE has committed $45 million to this problem through the CATALOG consortium and a 2024 FOA, and no commercial product performs in-flight anomaly confirmation.

Embodied AI

MESO TRL 4 July 27, 2026

RL-Guided Autonomous Robotic Ultrasound for Point-of-Care Diagnostics

Two independent systems have demonstrated autonomous ultrasound scanning matching or exceeding physician performance: FARUS tested on 19 patients with diagnostic-quality thyroid scans (Nature Communications, 2024), and auto-RUSS achieving 0.5% thyroid volume error versus CT gold standard (34.3 mL vs 34.48 mL, p=0.285), compared to 4.3–10.1% error from expert physicians (Frontiers in Robotics and AI, 2025). Sonographer vacancy rates reached 12.4% in 2025 while US exam volumes grew 55% in a decade. The robotic ultrasound systems market was valued at $1.52B in 2024 growing at 16.6% CAGR to $3.85B by 2032. Zero commercial products offer fully autonomous RL-guided diagnostic scanning; all existing robotic systems require human teleoperation. The integration gap between validated RL control, compliant robotic hardware, and clinical-grade imaging represents a first-mover opportunity with a $664M annual procedural revenue addressable market and FDA De Novo pathway precedent (IDx-DR, 2018).

Embodied AI

DECA TRL 4 July 20, 2026

RL-Guided Autonomous Robotic Disassembly for EV Battery and E-Waste Recycling

Five independent research groups (Iowa State/RAISE, Birmingham, Chinese Academy of Sciences, Fraunhofer IFF, ORNL/DOE) have demonstrated components of autonomous robotic disassembly since 2023, including vision-guided phone disassembly at 98.9% success and 120+ units per hour (arXiv, 2025), NeuroSymbolic EV battery bolt removal with 154.4% efficiency gain (Batteries, 2025), and AI-driven disassembly planning from a systematic review of 62 studies (Frontiers in Robotics and AI, 2025). The global e-waste management market is $77B growing to $120B by 2030 (Mordor Intelligence); EV battery recycling reaches $56.3B by 2031 (MarketsandMarkets). Current practice relies on manual disassembly (dangerous, slow) or mechanical shredding (destroys 20-40% of recoverable value). Zero commercial products offer general-purpose RL-guided autonomous disassembly. The integration gap between validated vision, RL planning, and robotic manipulation represents a first-mover opportunity in a $3.5B to $8.2B addressable US market with $3B+ in DOE funding signals for domestic battery materials supply chain.

Embodied AI

DECI TRL 4 July 13, 2026

Reinforcement Learning for Autonomous Robotic Bronchoscopy Navigation in Peripheral Lung Lesion Diagnosis

Five independent research groups (Zhejiang University, Harbin Institute of Technology, Shanghai Jiao Tong University, Harvard/BWH) have demonstrated AI-assisted or conditionally autonomous bronchoscope navigation since 2024, including a live porcine AI co-pilot (Nature Communications, 2024), 5G remote foreign body retrieval at 1,500 km (Science Robotics, 2025), and pure-vision RL navigation reaching eighth-generation airways at 80% success (arXiv, 2026). Lung cancer is the #1 cancer killer with 226,650 new US cases annually. Three FDA-cleared robotic bronchoscopy platforms (Ion, Monarch, Galaxy) represent a $750M market growing at 17.2% CAGR to $2.6B by 2033, yet all three require continuous physician control. Zero commercial products offer autonomous navigation. The gap between validated prototypes and a software autonomy layer on existing FDA-cleared hardware represents a first-mover opportunity in a $200M to $500M annually addressable US market with a 510(k) regulatory pathway.

Embodied AI

MESO TRL 4 June 15, 2026

AI-Guided Autonomous Robotic Thrombectomy for Acute Ischemic Stroke

Three independent demonstrations (Artedrone/Carvolix autonomous catheter navigation in porcine cerebral vasculature, King’s College London TD-MPC2 reinforcement learning achieving 65% success across 10 patient vasculatures at MICCAI 2025, and ETH Zurich magnetic microrobots performing vascular thrombectomy in large animals published in Science 2025) converge on a single clinical target: the 88% mechanical thrombectomy access gap identified by ARPA-H. Approximately 800,000 strokes occur annually in the US, yet only 12% of eligible patients receive thrombectomy due to a bottleneck of roughly 2,000 neurointerventionalists and severe rural access disparity (27.7% vs 69.5% urban). Zero commercial products provide autonomous neurovascular thrombectomy. The gap between validated prototypes and clinical deployment represents a first-mover opportunity in an $8.4B to $17.5B annually addressable US market with explicit federal funding signals from ARPA-H AIR and DARPA MASH.

Embodied AI

MICRO TRL 4 June 8, 2026

Magnetically Actuated Microrobots for Active Biofilm Eradication on Indwelling Medical Devices

Four independent research groups (UPenn, CUHK, ITMO University, Czech Academy of Sciences) have demonstrated magnetically actuated microrobots achieving 93.45% bacterial reduction in vivo (P<0.0001) and 100% biofilm eradication in catheter models. Healthcare-associated infections cause 72,000 to 90,000 US deaths annually, costing $28.4 to $45 billion. Zero commercial products provide active, controllable biofilm destruction on indwelling devices. Every published system relies on manual magnetic control with no autonomous AI navigation. The gap between validated prototypes and clinical deployment represents a first-mover opportunity in a $2.98B US addressable market across catheters, stents, and orthopedic implants.

Embodied AI

KILO TRL 5 June 1, 2026

Autonomous UAV Systems for Post-Blast Underground Mine Gas Monitoring and Safety Inspection

Two independent research groups (Lulea University of Technology and University of Manchester) have demonstrated field-validated prototypes achieving autonomous gas measurement 40 minutes post-blast at 700m depth and borehole-deployable drones navigating through 130mm apertures. MSHA recorded 10 miner fatalities in the first 10 weeks of 2025, triple the prior year rate. Zero commercial products combine autonomous GPS-denied navigation with integrated toxic gas sensing for post-blast re-entry assessment. Over 2,300 active underground mines worldwide lose 15 to 30% of productive time to conservative re-entry waiting periods. The gap between validated prototypes and mine-rated products represents a first-mover opportunity in a $500M to $1.1B market segment.

Embodied AI

MICRO TRL 4–5 May 18, 2026

Wireless Smart Stents for Autonomous In-Vivo Restenosis Monitoring

Six independent research groups (UCLA, Georgia Tech, University of Pittsburgh, UBC, University of Glasgow, Chonnam National University) have demonstrated functional wireless smart stent prototypes across three animal models, achieving wireless communication ranges up to 50 cm and real-time thrombus detection with p<0.001 discriminatory power. In-stent restenosis affects 5–15% of the 600,000 US patients receiving coronary stents annually, currently detected only through invasive angiography costing $1,708–$3,312 per procedure. Zero commercial wireless restenosis-monitoring stents exist despite $12.1M in combined ARPA-H and EU funding awarded to smart stent startups in 2024. Three gaps separate published prototypes from clinical deployment: AI-driven adaptive baseline learning, scalable sensor-integrated stent manufacturing, and modular FDA regulatory architecture.

Embodied AI

MICRO TRL 4 May 11, 2026

Autonomous Microrobotic Swarms for Aquatic Microplastic Remediation

Three independent research trajectories—magnetotactic bacteria biobots (Song 2025), photocatalytic Ag@Bi₂WO₆/Fe₃O₄ composites (Wang 2024), and gallium-based liquid metal LiquidBots (Wu 2025)—have demonstrated microplastic capture efficiencies of 89–96% within 2–4 hours in laboratory settings. Microplastics contaminate 94% of US tap water and pass through conventional wastewater treatment unchanged. No existing water treatment technology actively targets particles below 100 μm. The gap between validated microrobot architectures and deployable water treatment modules represents a first-mover opportunity in a $2.6B US addressable market across 16,000 publicly owned treatment works.

Embodied AI

DECA TRL 4 May 4, 2026

Autonomous Fall-Prevention Mobile Robotics for Elderly Independent Living

Three independent research trajectories (MIT CSAIL E-BAR platform, multiple sensor-fusion teams, wearable airbag validation studies) have demonstrated component technologies achieving full body-weight support through 38cm doorways, fall prediction AUC of 0.98, and 81% hip fracture reduction from airbag protection. Falls kill 43,020 Americans annually and cost $80 billion in healthcare spending. Zero commercial products combine mobile autonomous following, physical support, predictive fall detection, and airbag deployment. The gap between validated components and an integrated deployable system represents a first-mover opportunity in a $3.38B–$9.85B market.

Embodied AI

KILO TRL 5 April 27, 2026

Autonomous Precision Pollination Robots for Tree Fruit Orchards

Three independent research groups (Washington State University, Zhejiang Academy of Agricultural Sciences, West Virginia University) have demonstrated field-validated prototypes achieving 34.8–99.3% pollination success rates in commercial orchards. US managed honeybee colony losses reached 55.6% in 2024–2025 — the highest ever recorded. Zero commercial products exist for targeted liquid pollination of tree fruit orchards. The gap between validated research prototypes and fleet-deployable autonomous systems represents a first-mover opportunity in a $480–580M market.

Embodied AI

DECI TRL 4 April 13, 2026

Autonomous Closed-Loop Bioelectronic Wound Dressings for Chronic Wound Therapy

Four independent research groups (Caltech, Stanford, UC Santa Cruz, University of Arizona) have demonstrated field-validated prototypes achieving closed-loop healing acceleration and ML-based wound classification comparable to expert clinicians in a 20-patient human trial. Zero commercial products combine real-time wound biomarker monitoring with autonomous closed-loop drug delivery. The US has 8.2 million Medicare beneficiaries with chronic wounds generating $28 billion in annual spending.

Embodied AI

HECTO TRL 4 April 6, 2026

DRL-Guided Autonomous Structural Crack Inspection for Steel Bridge Fatigue Prognosis

Three independent groups have demonstrated field-validated prototypes achieving 59.6% inspection time reduction, 85% autonomous crack detection rates, and 3x faster data collection. Zero commercial products combine DRL-guided autonomous crack exploration with finite-element digital twin fatigue life prediction. The US has 617,000 bridges requiring biennial inspection, with 42,067 rated structurally deficient. The Infrastructure Investment and Jobs Act committed $40 billion to bridge investment.

Embodied AI

CENTI TRL 4 March 30, 2026

Autonomous In-Pipe Robots for Drinking Water Infrastructure Condition Assessment

Four independent systems (Pipebots at Sheffield, MIT Mechatronics, SubMerge in Netherlands, MotBot at Motmot Inc.) have demonstrated tetherless in-pipe robots in water distribution networks. Zero commercial products exist for autonomous inspection of live pressurized drinking water mains. The gap between demonstrated prototypes and utility-scale deployment — multi-sensor fusion, miniaturized manufacturing, and utility data integration — represents a first-mover opportunity in a $1.37B annual recurring US market.

Embodied AI

DECA TRL 5 March 23, 2026

Self-Driving Laboratories for Autonomous Materials Discovery

Three independent systems (A-Lab at Berkeley, Polybot at Argonne, MINERVA at BAM) have demonstrated continuous autonomous materials synthesis. Zero commercial SDL products exist. The DOE committed $320M through the Genesis Mission (December 2025). The gap between validated prototypes and deployable products — scalable active learning algorithms, modular hardware integration, and manufacturing scale-up — represents a first-mover opportunity in a $5.7B US market.

Embodied AI

MESO TRL 5 March 16, 2026

RL-Adaptive Rehabilitation Exoskeletons for Post-Stroke Gait Recovery

Four independent groups have demonstrated 24% metabolic cost reductions using reinforcement learning on hip and ankle exoskeletons (Nature 2022, 2024). Zero commercial rehabilitation exoskeletons use RL-adaptive control — all FDA-cleared devices operate with fixed, pre-programmed gait trajectories. The gap between demonstrated RL performance and clinical deployment represents a first-mover opportunity in a device class where Medicare reimbursement already exists.

Embodied AI

MESO TRL 4 March 14, 2026

Autonomous Soft Tissue Surgical Systems with Learned Dexterity

All 53 FDA-cleared surgical robots operate at Level 1-3 autonomy — the surgeon performs every action. Recent demonstrations of Level 4 autonomy in soft tissue (Johns Hopkins STAR, SRT-H) show that imitation learning from surgical video can achieve superhuman precision in specific procedures. No commercial entity is pursuing Level 4 surgical autonomy.

Embodied AI

MICRO TRL 4 March 13, 2026

Closed-Loop Adaptive Bioelectronic Implants for Inflammatory Disease

Autoimmune and chronic inflammatory diseases affect over 24 million Americans. Current immunosuppressive therapies carry severe side effects and lose efficacy over time. Bioelectronic medicine offers a fundamentally different approach: precisely timed electrical stimulation of the vagus nerve to modulate the cholinergic anti-inflammatory pathway.

Embodied AI

MICRO TRL 4 March 13, 2026

Magnetically Guided Microrobots for Targeted Intravascular Drug Delivery

Systemic chemotherapy distributes cytotoxic agents throughout the entire body. Fewer than 1% of administered drug reaches the tumor site; the remainder damages healthy tissue, producing dose-limiting toxicities including neutropenia, nephrotoxicity, cardiotoxicity, and peripheral neuropathy.

Embodied AI

Published Research

Open-source simulation platforms

Reinforcement learning environments for safety-critical medical applications. All packages available on PyPI with comprehensive test suites.

1

AnestheSim — RL environments for automated anesthesia drug dosing (109 tests)

2

NeuroSim — RL platform for brain-computer interfaces (158 tests)

3

VascularSim — Microbot navigation in patient-derived vascular networks (139 tests)

4

PeptideGym — RL environments for therapeutic peptide design (125 tests)

5

OncoSim — RL for radiation therapy treatment planning (141 tests)

6

GlucoSim — RL for glucose management in Type 1 diabetes (117 tests)

7

VentiSim — RL for mechanical ventilation control (230 tests)

8

CardioSim — RL for cardiac electrophysiology (134 tests)

Research originally published by Smart Technology Investments Research Institute (smarttechinvest.com/research). Licensed to HHA Applied Research Institute under a non-exclusive research license for R&D and grant pursuit. Commercial exploitation rights retained by STI.