{
  "id": "autonomous-systems-vehicles",
  "label": "Autonomous Systems & Vehicles",
  "description": "AI systems that navigate and operate in the physical world across road, air, marine, rail, and space. Heavily leading-edge: ADAS and warehouse robotics are good practice, but full autonomy in open environments remains elusive. Evenly split between advancing and stalled — regulatory frameworks and edge-case safety are the binding constraints. One of the largest domains with 22 practices spanning a wide maturity range.",
  "icon": "🚗",
  "filters": [
    "moving"
  ],
  "hasSummary": true,
  "hasExecSummary": true,
  "practiceCount": 21,
  "evidenceCount": 4045,
  "practices": [
    {
      "slug": "adas-driver-assistance-and-emergency-intervention",
      "name": "ADAS — driver assistance & emergency intervention",
      "tier": "good-practice",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-powered advanced driver assistance including lane keeping, adaptive cruise, parking assist, and autonomous emergency braking and evasion. Includes sensor fusion for driver support and collision avoidance; distinct from autonomous vehicles which drive without human supervision. Scope covers ML/neural-net-based perception and decision systems; traditional radar-only cruise control and non-ML mechanical safety features are out of scope.",
      "evidenceCount": 213
    },
    {
      "slug": "autonomous-buses-and-public-transit-shuttles",
      "name": "Autonomous buses & public transit shuttles",
      "tier": "leading-edge",
      "trend": "slowing",
      "blockerType": null,
      "description": "Self-driving buses and shuttles operating on fixed or flexible routes in public transit systems. Includes campus shuttles and fixed-route transit; distinct from robotaxis which provide on-demand rather than scheduled service. Scope covers ML-perception-based autonomous buses navigating real traffic; deterministic guideway transit and fixed-track people movers without ML are out of scope.",
      "evidenceCount": 207
    },
    {
      "slug": "autonomous-marine-vessels",
      "name": "Autonomous marine vessels",
      "tier": "leading-edge",
      "trend": "accelerating",
      "blockerType": null,
      "description": "AI-navigated autonomous cargo ships, ferries, and water taxis operating in coastal and open-water environments. Includes autonomous collision avoidance and route planning; distinct from port operations which manage docking and cargo handling rather than vessel navigation.",
      "evidenceCount": 188
    },
    {
      "slug": "autonomous-mining-and-off-road-vehicles",
      "name": "Autonomous mining & off-road vehicles",
      "tier": "good-practice",
      "trend": "steady",
      "blockerType": null,
      "description": "Self-driving vehicles operating in mining, quarrying, and off-road environments including haul trucks and loaders. Includes autonomous haulage systems and drill automation; distinct from autonomous earthmoving in construction which operates on building sites rather than mines. Scope covers ML-perception-based autonomy and AI-driven path planning; GPS-waypoint-following haulage without ML perception is out of scope.",
      "evidenceCount": 194
    },
    {
      "slug": "autonomous-passenger-vehicles-highway-l3",
      "name": "Autonomous passenger vehicles — highway (L3)",
      "tier": "leading-edge",
      "trend": "slowing",
      "blockerType": null,
      "description": "Self-driving cars capable of highway driving with human driver available as fallback (SAE Level 3). Includes traffic jam pilot and highway autopilot; distinct from urban L4 which operates in complex city environments without requiring human fallback. Scope covers ML/AI-driven approaches; prior deterministic or rules-based automation is out of scope.",
      "evidenceCount": 170
    },
    {
      "slug": "autonomous-passenger-vehicles-urban-l4",
      "name": "Autonomous passenger vehicles — urban (L4)",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "Self-driving cars operating in defined urban areas without requiring human fallback (SAE Level 4). Includes geofenced robotaxi services and urban autonomous operation; distinct from L5 which operates in all conditions rather than defined operational domains. Scope covers ML/AI-driven approaches; prior deterministic or rules-based automation is out of scope.",
      "evidenceCount": 189
    },
    {
      "slug": "autonomous-port-operations-and-container-handling",
      "name": "Autonomous port operations & container handling",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-controlled automated port systems for container handling, stacking, and vessel loading/unloading. Includes automated stacking cranes and AGV yard management; distinct from autonomous shipping which navigates at sea rather than operating in port. Scope covers ML-based perception, planning, and optimisation for port vehicles and cranes; pre-programmed crane automation and fixed-path AGVs without ML are out of scope.",
      "evidenceCount": 193
    },
    {
      "slug": "autonomous-rail-operations-and-scheduling",
      "name": "Autonomous rail operations & scheduling",
      "tier": "leading-edge",
      "trend": "accelerating",
      "blockerType": null,
      "description": "AI that automates train operation and optimises rail network scheduling and capacity allocation. Includes GoA 3/4 automated train control and predictive schedule optimisation; distinct from rail infrastructure inspection which monitors physical assets rather than operating services. Scope begins at AI-based ATO optimisation and ML-driven predictive scheduling; deterministic GoA-rated automation without ML components is out of scope.",
      "evidenceCount": 200
    },
    {
      "slug": "autonomous-satellite-operations-and-collision-avoidance",
      "name": "Autonomous satellite operations & collision avoidance",
      "tier": "leading-edge",
      "trend": "accelerating",
      "blockerType": null,
      "description": "AI systems managing satellite constellations, orbital manoeuvres, and collision avoidance with space debris. Includes automated conjunction assessment and manoeuvre planning; distinct from planetary exploration which operates on celestial surfaces rather than in orbit.",
      "evidenceCount": 204
    },
    {
      "slug": "autonomous-trucking-highway-hub-to-hub",
      "name": "Autonomous trucking — highway hub-to-hub",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "Self-driving trucks operating autonomously on highways between logistics hubs with human drivers for first/last mile. Includes platooning and transfer hub operations; distinct from last-mile delivery which navigates local roads to final destinations. Scope covers ML-perception-based autonomous driving and AI-enabled platooning; conventional cruise control and non-ML driver-assist systems are out of scope.",
      "evidenceCount": 182
    },
    {
      "slug": "autonomous-trucking-last-mile-delivery",
      "name": "Autonomous trucking — last-mile delivery",
      "tier": "bleeding-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "Self-driving delivery trucks navigating local roads for final delivery to customers and businesses. Includes neighbourhood navigation and delivery interaction; distinct from hub-to-hub which uses simpler highway environments. Scope covers ML-perception-based autonomous navigation and AI-driven delivery planning; manually driven or remote-controlled delivery vehicles without ML are out of scope.",
      "evidenceCount": 170
    },
    {
      "slug": "counter-drone-detection-and-mitigation",
      "name": "Counter-drone detection & mitigation",
      "tier": "leading-edge",
      "trend": "accelerating",
      "blockerType": null,
      "description": "AI systems that detect, track, classify, and mitigate unauthorised drone activity in restricted airspace. Includes RF detection, radar tracking, and electronic countermeasures; distinct from air traffic management which coordinates authorised rather than countering unauthorised flights. Scope covers AI/ML-based detection, classification, and tracking; conventional RF scanning and radar without ML processing is out of scope.",
      "evidenceCount": 206
    },
    {
      "slug": "drone-delivery-rural-and-suburban",
      "name": "Drone delivery — rural & suburban",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "Autonomous drone delivery services operating in rural and suburban areas with lower airspace complexity. Includes medical supply delivery and remote area logistics; distinct from urban drone delivery which faces higher density and regulatory challenges. Scope covers ML/AI-driven approaches; prior deterministic or rules-based automation is out of scope.",
      "evidenceCount": 204
    },
    {
      "slug": "drone-delivery-urban",
      "name": "Drone delivery — urban",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "Autonomous drone delivery in dense urban environments requiring sophisticated navigation and airspace management. Includes rooftop and balcony delivery and urban airspace deconfliction; distinct from rural delivery which operates in simpler airspace. Scope covers ML/AI-driven approaches; prior deterministic or rules-based automation is out of scope.",
      "evidenceCount": 164
    },
    {
      "slug": "drone-inspection-infrastructure-energy-and-agriculture",
      "name": "Drone inspection — infrastructure, energy & agriculture",
      "tier": "good-practice",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-guided drones that autonomously inspect infrastructure, energy assets, and agricultural land using visual and multispectral sensors. Includes automated defect detection and crop health mapping; distinct from drone delivery which transports goods rather than gathering data.",
      "evidenceCount": 191
    },
    {
      "slug": "fleet-management-route-optimisation-and-v2x",
      "name": "Fleet management, route optimisation & V2X",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI that optimises fleet routing, manages vehicle coordination, and enables vehicle-to-everything communication for connected transport. Includes real-time route optimisation and cooperative driving coordination; distinct from autonomous vehicles which navigate independently rather than as coordinated fleets. Scope covers ML-driven dynamic routing and AI-based coordination; classical operations-research routing without ML is out of scope.",
      "evidenceCount": 201
    },
    {
      "slug": "hazardous-and-extreme-environment-exploration",
      "name": "Hazardous & extreme environment exploration",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-powered robotic systems that explore and inspect deep-sea, contaminated, high-radiation, and other hazardous environments unsafe for humans. Includes autonomous deep-sea mining and nuclear facility inspection; distinct from underwater inspection which targets infrastructure rather than exploration.",
      "evidenceCount": 187
    },
    {
      "slug": "marine-inspection-and-environmental-monitoring",
      "name": "Marine inspection & environmental monitoring",
      "tier": "leading-edge",
      "trend": "accelerating",
      "blockerType": null,
      "description": "AI-powered underwater vehicles and sensors that inspect subsea infrastructure and monitor marine environmental conditions. Includes autonomous pipeline inspection and ocean condition monitoring; distinct from autonomous vessels which navigate on the surface.",
      "evidenceCount": 181
    },
    {
      "slug": "planetary-exploration-robotics",
      "name": "Planetary exploration robotics",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-powered rovers and exploration robots operating autonomously on other planets and celestial bodies. Includes autonomous path planning and science target selection; distinct from satellite operations which operate in orbit rather than on surfaces.",
      "evidenceCount": 228
    },
    {
      "slug": "rail-infrastructure-inspection",
      "name": "Rail infrastructure inspection",
      "tier": "leading-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI-powered inspection of rail tracks, signals, and infrastructure using vision, sensors, and drone systems. Includes rail defect detection and geometry measurement; distinct from structural health monitoring which targets buildings and bridges rather than rail.",
      "evidenceCount": 186
    },
    {
      "slug": "urban-air-mobility-and-air-traffic-management",
      "name": "Urban air mobility & air traffic management",
      "tier": "bleeding-edge",
      "trend": "steady",
      "blockerType": null,
      "description": "AI systems for eVTOL/air taxi operations and autonomous management of urban and general air traffic. Includes AI flight planning and autonomous separation management; distinct from drone operations which handle smaller unmanned aircraft.",
      "evidenceCount": 187
    }
  ],
  "summary": "## Where AI Stands in Autonomous Systems & Vehicles\n\nAutonomy in the physical world now divides along one line. It thrives where a single owner controls the environment and carries the consequences, and it stalls where it has to share roads, airspace, liability or budgets with others. The first group is already at industrial scale. China had roughly 8,500 autonomous mining trucks in service by mid-2026, after delivering about 4,100 in 2025 alone. Rio Tinto's Pilbara haul fleet is reported to be about 90% driverless, and Epiroc has supplied more than 3,900 driverless underground machines. SpaceX's Starlink carries out around 1,000 automated collision-avoidance manoeuvres a day. Driverless GoA4 signalling is now the default specification for new metros, and Thessaloniki opened another driverless branch at the end of August. On the water, Avikus's HiNAS Control runs across 40 HMM and 14 KMTC vessels under DNV type approval. Defence buyers have announced more than $53bn of counter-drone contracts this year. These deployments differ in almost every technical respect. What they share is a single owner that writes the operating rules, absorbs the risk and does not need the public's permission.\n\nThe second group is where the headlines are, and its commercial footing is much weaker. Waymo now runs more than 4,000 driverless vehicles across 14 US cities and covers about 4m miles a week. Aurora has run heavy trucks on Texas interstates with no observer on board since July, and expects about 200 by year-end. Zipline, Wing and Amazon are expanding suburban drone networks into hundreds of towns. All of it is real service, and none of it yet pays its way. Aurora lost $270m in the second quarter on $2m of revenue. Pony.ai, whose chief executive calls the technology \"pretty much solved\", posted a $65.7m operating loss. Consumer highway autonomy shows the same split along geographic lines. Western carmakers have largely retreated to supervised driver assistance, while China issues L3 product admissions under a binding national standard. That still means only a few dozen cars on restricted roads, not a mass market.\n\nThe domain now looks less like an engineering race than an institutional one. The perception and planning stacks broadly work inside their operating envelopes. What limits them is who carries liability, who coordinates shared airspace and traffic, whether the software vendor will still be there in three years, and whether the operator's data is good enough to run the models. Independent measurement is also becoming less kind to vendor claims. A 56-port study found no throughput gain from terminal automation that could be told apart from a placebo. A benchmark of AI track-defect detection on Indian railway data caught fewer than half the burned rails in its test set. Consultancy surveys of rail operators found most AI initiatives stuck at pilot stage. Put plainly, the technology delivers where it has been given a controlled environment to work in. Beyond that, the remaining problems belong to regulators, insurers and finance directors more than to engineers.\n\n## What's New, 2026-08-31 to 2026-09-28\n\nThe most telling news this month was about services shutting down, and none of the closures was about safety. Jacksonville's NAVI, one of the more established US autonomous transit shuttles, reverts to manual driving in January 2027 after $39m of budget cuts and the expiry of its software licence. Monheim am Rhein, one of Germany's first autonomous bus lines, suspended service after EasyMile withdrew support, following EasyMile's exit from passenger transport earlier this year. NASA was reported to have no new Mars lander or rover plans for the first time in more than 30 years, and is turning to AeroVironment helicopters instead. A fatality halted BHP's Escondida, the world's largest copper mine and a major autonomous-haulage site, although the death has not been attributed to the autonomy system. Expansion carried on regardless. Waymo opened in San Diego, Tampa and Denver. Tesla's Cybercab, which has no steering wheel, began charging fares in Austin and drew an NHTSA audit query within hours. Volvo Autonomous Solutions set a driverless trucking launch for the first quarter of 2027. Zipline signed an eight-year national contract in Côte d'Ivoire and is reportedly in talks to raise about $1bn at a valuation of around $20bn. Amazon plans to take Prime Air to nearly 500 US towns. Meituan's drone corridor on Shanghai's West Bund, built with McDonald's, pushed it past 1m paid orders.\n\nRegulation and independent evidence mostly pulled the other way. The FAA's $875m AI traffic-flow system, SMART, went live at three Washington-area airports, but an outage marred the launch and the controllers' union was left out of its design. The FAA's Part 108 rule for routine beyond-line-of-sight drone flight has not taken effect: it is still under review at the White House regulatory office (OIRA), despite operators setting launch dates as if it had. The federal TraCSS space-traffic service remains in evaluation mode, while SpaceX's free Stargaze platform signed up more than 30 satellite operators in a week. The IMO's new code for autonomous ships (the MASS Code) turns out to grant permission to operate but no liability protection. A 100% Section 232 tariff on thermal-imaging drones and docks took effect in the US on 3 September. Australia's crash-test body, ANCAP, will begin penalising intrusive or badly calibrated driver assistance. Tesla's own federal crash filings show its driver assistance engaged in fatal crashes in New Jersey and Texas, with the software version and operating-area details redacted.\n\n## Key Tensions\n\n- **Operators rent autonomy from vendors who can walk away.** Jacksonville and Monheim lost working services because a software licence lapsed or a supplier left the market, not because anything went wrong on the road. Hyundai pushed its own driver-assist stack back to 2029 and will ship Nvidia's Hyperion 10 platform in 2028 instead. Deutsche Bahn's consortium notes that no autonomous vehicle has yet completed full German type approval. Anyone buying autonomy as a service is exposed to the vendor's balance sheet, whatever the technology's merits.\n\n- **Permission is granted, liability is not.** The IMO's MASS Code authorises autonomous ship operations but gives no liability protection. It excludes passenger ships and cargo ships under 500 gross tonnes, and P&I clubs still do not recognise remote-centre staff as seafarers. US hub-to-hub trucking has no national safety framework, while a patchwork of state laws grows. In aviation, both a 33-partner European consortium and the director general of France's civil aviation school, ENAC, judge airspace integration unready, even as individual aircraft move through certification.\n\n- **Vendor metrics against independent measurement.** Kodiak's 93% readiness figure is self-reported. A drone-inspection vendor itself says about 40% of utility inspection imagery is unusable. Tesla redacts key fields in its own crash filings. Independent work is harsher. A difference-in-differences study of 56 ports found no statistically distinguishable throughput gain from automation. An independent benchmark of AI track-defect detection found 45.5% recall on burned rails. McKinsey reports 88% of shippers satisfied with transport AI, but a separate fleet survey finds 74% have no formal AI strategy.\n\n- **Scale is running ahead of unit economics.** Aurora prices its planned Driver-as-a-Service above $0.85 a mile, against about $1.18 for a human driver, while losing $270m a quarter. Waabi's own chief operating officer warns that human-driven transfer legs add hundreds of dollars per load. Waymo's newest vehicles are reported to cost up to about $125,000 each. In rural drone delivery, aid cuts now threaten the Rwandan clinics that Zipline's network depends on, a reminder that a working logistics network does not make a viable service without the institutions at the other end.\n\n- **Automated thresholds are becoming decisions without clear accountability.** Critics argue that Starlink's 3x10^-7 collision-probability threshold now works as de facto manoeuvre authority, with no clear accountability, across 355,000 manoeuvres a year. The FAA built SMART without its controllers' union in the room. A Delft study of 60 drivers found driver-assistance use degraded manual skills and increased risk-taking. Peer-reviewed reviews of highway L3 still flag unresolved takeover and misuse risk. In each case, how far to trust the automation is being settled in practice before anyone has settled it in governance.\n\n## Top 10 Evidence Items\n\n1. **China's autonomous mining trucks move from demonstration projects to 100-unit commercial fleets (Taiyuan Energy Expo / China International Mining Conference)** (news-coverage) — Anchors the briefing's controlled-environment thesis: single-owner mining fleets scaling into the thousands while public-road autonomy stalls. https://eu.36kr.com/en/p/3988167128808195\n2. **Jacksonville Transportation Authority pauses NAVI autonomous shuttle service amid $39M budget cuts and software licence expiry** (news-coverage) — Shows the vendor-dependency tension directly — a working shuttle service killed by a lapsed software licence and budget cuts, not a safety failure. https://www.news4jax.com/news/local/2026/09/17/jta-to-cut-39-million-from-fy-2027-budget-suspend-skyway-service-in-november/\n3. **Monheim am Rhein suspends autonomous bus line A01 after EasyMile ends software support** (news-coverage) — A second, independent case of the same pattern as Jacksonville, confirming vendor withdrawal rather than technical failure as the recurring cause of shutdowns. https://www.busplaner.de/de/news/easymile-streicht-support-monheim-stellt-autonome-busline-ein-121241.html\n4. **French maritime law firm: MASS Code secures authorisation to operate but not liability** (opinion) — Gives the legal substance behind the 'permission granted, liability not' tension — the MASS Code's actual gaps in coverage and seafarer status. https://www.guyader-avocat.fr/en/autonomous-ships-imo-mass-code-liability-2026/\n5. **The Space Review: a collision probability is not permission to manoeuvre (critique of Starlink's automated 3×10⁻⁷ threshold)** (opinion) — Illustrates automated thresholds quietly becoming manoeuvre authority without governance, the clearest case of the accountability-vacuum tension. https://www.thespacereview.com/article/5312/1\n6. **Staggered difference-in-differences study finds no statistically distinguishable throughput gain from terminal automation across 56 ports** (research-paper) — The strongest independent counter-evidence to vendor automation claims, showing port throughput gains are statistically indistinguishable from placebo. https://jemsjournal.org/articles/automation-and-container-throughput-heterogeneous-evidence-from-a-staggered-difference-in-differences-design/jems.2026.77527\n7. **Independent CNN benchmark on Indian railway track defects finds overfitting and weak burned-rail recall** (research-paper) — A concrete independent benchmark undercutting AI inspection hype, catching under half the safety-critical defects it was meant to find. https://www.mdpi.com/2813-477X/4/3/27\n8. **Aurora CEO at Goldman Sachs conference: observer-free driverless operations, 200 trucks by year-end and Driver-as-a-Service pricing** (conference-talk) — Captures the unit-economics gap at the centre of the second theme: driverless trucking scaling while still losing hundreds of millions against thin revenue. https://www.investing.com/news/transcripts/aurora-at-goldman-sachs-conference-driverless-trucking-shifts-to-scale-93CH-4897020\n9. **Fatal accident halts BHP's Escondida, a mine running autonomous haul trucks** (news-coverage) — Tests the 'controlled environment' thesis at its most industrial site — the world's largest copper mine, where autonomy's safety case is now under real scrutiny. https://www.geomechanics.io/news/article/bhps-escondida-restart-market-balance-safety-and-project-lens-for-mine-planners\n10. **Tesla Cybercab paid launch in Austin prompts NHTSA Audit Query AQ26002 on self-certification** (industry-report) — Shows regulatory institutions, not engineering, as the live constraint on L4/L5 rollout, with a federal audit opened within hours of a driverless launch. https://www.cognitivemarketresearch.com/articles/teslas-robotaxi-launched-hours-later-the-feds-hit-back",
  "execSummary": "**The headline:** Autonomy works where one owner controls the site. This month, two working driverless shuttle services were suspended or scheduled to end over budget cuts, a lapsed software license and a vendor's exit, not safety. The bigger risk now is the supplier, not the technology.\n\n### The Picture\n\nWhere a single company controls the environment and carries the risk, autonomy already runs at industrial scale. China has roughly 8,500 autonomous mining trucks in service, Rio Tinto's Pilbara haul fleet is about 90% driverless, and driverless signaling is now the default specification for new metro lines. Services on shared roads and in shared airspace are real but still lose money. That includes Waymo's 4,000-plus robotaxis in 14 US cities, Aurora's driverless trucks in Texas and suburban drone delivery. Most companies will buy autonomy as a service rather than build it, so the deciding questions are vendor staying power, liability and data quality, not whether the software can drive. If you run a mine, yard, port or depot, your peers are already operating this. If you depend on public roads or airspace, waiting is still a defensible choice.\n\n### This Fortnight\n\n- **Two established autonomous shuttle services are ending for commercial reasons, not safety ones.** Jacksonville's NAVI shuttle returns to human drivers in January 2027 after $39M of budget cuts and the expiry of its software license. Monheim am Rhein in Germany suspended its line after the vendor, EasyMile, withdrew support. Anyone signing an autonomy contract should now treat license continuity, exit terms and vendor viability as core terms of the deal.\n\n- **Robotaxi expansion kept going while regulators moved closer to purpose-built vehicles.** Waymo opened paid driverless service in San Diego, Tampa and Denver. Tesla's Cybercab, which has no steering wheel, began charging fares in Austin and drew an audit query from the federal auto safety regulator (NHTSA) within hours. Expect more scrutiny of vehicles that cannot be handed back to a human driver.\n\n- **The FAA's $875M AI air traffic flow system went live at three Washington-area airports, and the launch was marred by an outage.** The controllers' union was left out of the system's design. Where automation changes how people work, leaving the people who run it out of the design is an operational risk as well as a labor-relations one.\n\n- **A 100% US tariff on thermal-imaging drones and their docking stations took effect on September 3.** These are the workhorses of utility, solar and pipeline inspection. Operators should expect higher hardware costs at renewal and should re-price drone inspection programs before committing next year's budget.\n\n- **Independent research undercut vendor claims about automation payoffs.** A peer-reviewed study of 56 ports found no throughput gain from terminal automation that could be told apart from a placebo, and a separate benchmark of AI rail-defect detection missed most burned rails. Ask for independently measured results, not vendor case studies, before approving capital spending.\n\n### Coming Up\n\n- **Driverless long-haul trucking moves from a single frontrunner to several suppliers during 2027.** Aurora expects about 200 trucks without anyone on board by year-end, Volvo Autonomous Solutions targets a driverless launch in the first quarter of 2027, and Aurora plans to price its service above $0.85 a mile against roughly $1.18 for a human driver. Shippers on Sun Belt freight lanes should price a pilot lane now and track whether the human-driven legs at each end of the route wipe out the savings.\n\n- **The FAA rule allowing routine drone flights beyond the pilot's line of sight is still under White House review.** Operators have set launch dates as if the rule were final, and Amazon plans to take drone delivery to nearly 500 US towns while residents in Richardson, Texas, protest the noise. Don't tie drone delivery or inspection plans to a go-live date until the rule is published, and check local ordinances on siting.\n\n- **China's binding safety standard for eyes-off highway driving takes effect in July 2027, with carmakers liable while the system is engaged.** Western carmakers have mostly retreated to supervised driver assistance, and Hyundai has pushed its own driving software back to 2029, relying on Nvidia's platform in the meantime. Fleet buyers should not plan on eyes-off highway driving from Western brands this decade, and should check how dependent their vehicle suppliers are on a single chip platform.\n\n### What's Hard About This\n\n- **You rent autonomy from vendors who can walk away.** Jacksonville and Monheim lost working services because a license lapsed or a supplier left the market, and Deutsche Bahn notes that no autonomous vehicle has yet completed full German type approval. Better software does not protect you from a supplier's balance sheet; contract terms and fallback plans do.\n\n- **Regulators grant permission to operate, but not protection from liability.** The International Maritime Organization's new code for autonomous ships authorizes operations without shielding operators from liability, and shipowners' mutual insurers still do not recognize remote-control-center staff as seafarers. US driverless trucking has no national safety framework, only a growing patchwork of state laws. Until liability is settled, expect insurers and legal teams, not engineers, to set the pace.\n\n- **Scale is running well ahead of the economics.** Aurora lost $270M in the second quarter on $2M of revenue, and every major robotaxi operator is still losing money. Today's service prices are partly subsidized by investors, so budget for the possibility that they rise, or that your provider does not last.",
  "headline": "Autonomy works where one owner controls the site. This month, two working driverless shuttle services were suspended or scheduled to end over budget cuts, a lapsed software license and a vendor's exit, not safety. The bigger risk now is the supplier, not the technology.",
  "execSummarySections": [
    {
      "id": "the-picture",
      "title": "The Picture",
      "body": "Where a single company controls the environment and carries the risk, autonomy already runs at industrial scale. China has roughly 8,500 autonomous mining trucks in service, Rio Tinto's Pilbara haul fleet is about 90% driverless, and driverless signaling is now the default specification for new metro lines. Services on shared roads and in shared airspace are real but still lose money. That includes Waymo's 4,000-plus robotaxis in 14 US cities, Aurora's driverless trucks in Texas and suburban drone delivery. Most companies will buy autonomy as a service rather than build it, so the deciding questions are vendor staying power, liability and data quality, not whether the software can drive. If you run a mine, yard, port or depot, your peers are already operating this. If you depend on public roads or airspace, waiting is still a defensible choice."
    },
    {
      "id": "this-fortnight",
      "title": "This Fortnight",
      "body": "- **Two established autonomous shuttle services are ending for commercial reasons, not safety ones.** Jacksonville's NAVI shuttle returns to human drivers in January 2027 after $39M of budget cuts and the expiry of its software license. Monheim am Rhein in Germany suspended its line after the vendor, EasyMile, withdrew support. Anyone signing an autonomy contract should now treat license continuity, exit terms and vendor viability as core terms of the deal.\n\n- **Robotaxi expansion kept going while regulators moved closer to purpose-built vehicles.** Waymo opened paid driverless service in San Diego, Tampa and Denver. Tesla's Cybercab, which has no steering wheel, began charging fares in Austin and drew an audit query from the federal auto safety regulator (NHTSA) within hours. Expect more scrutiny of vehicles that cannot be handed back to a human driver.\n\n- **The FAA's $875M AI air traffic flow system went live at three Washington-area airports, and the launch was marred by an outage.** The controllers' union was left out of the system's design. Where automation changes how people work, leaving the people who run it out of the design is an operational risk as well as a labor-relations one.\n\n- **A 100% US tariff on thermal-imaging drones and their docking stations took effect on September 3.** These are the workhorses of utility, solar and pipeline inspection. Operators should expect higher hardware costs at renewal and should re-price drone inspection programs before committing next year's budget.\n\n- **Independent research undercut vendor claims about automation payoffs.** A peer-reviewed study of 56 ports found no throughput gain from terminal automation that could be told apart from a placebo, and a separate benchmark of AI rail-defect detection missed most burned rails. Ask for independently measured results, not vendor case studies, before approving capital spending."
    },
    {
      "id": "coming-up",
      "title": "Coming Up",
      "body": "- **Driverless long-haul trucking moves from a single frontrunner to several suppliers during 2027.** Aurora expects about 200 trucks without anyone on board by year-end, Volvo Autonomous Solutions targets a driverless launch in the first quarter of 2027, and Aurora plans to price its service above $0.85 a mile against roughly $1.18 for a human driver. Shippers on Sun Belt freight lanes should price a pilot lane now and track whether the human-driven legs at each end of the route wipe out the savings.\n\n- **The FAA rule allowing routine drone flights beyond the pilot's line of sight is still under White House review.** Operators have set launch dates as if the rule were final, and Amazon plans to take drone delivery to nearly 500 US towns while residents in Richardson, Texas, protest the noise. Don't tie drone delivery or inspection plans to a go-live date until the rule is published, and check local ordinances on siting.\n\n- **China's binding safety standard for eyes-off highway driving takes effect in July 2027, with carmakers liable while the system is engaged.** Western carmakers have mostly retreated to supervised driver assistance, and Hyundai has pushed its own driving software back to 2029, relying on Nvidia's platform in the meantime. Fleet buyers should not plan on eyes-off highway driving from Western brands this decade, and should check how dependent their vehicle suppliers are on a single chip platform."
    },
    {
      "id": "whats-hard-about-this",
      "title": "What's Hard About This",
      "body": "- **You rent autonomy from vendors who can walk away.** Jacksonville and Monheim lost working services because a license lapsed or a supplier left the market, and Deutsche Bahn notes that no autonomous vehicle has yet completed full German type approval. Better software does not protect you from a supplier's balance sheet; contract terms and fallback plans do.\n\n- **Regulators grant permission to operate, but not protection from liability.** The International Maritime Organization's new code for autonomous ships authorizes operations without shielding operators from liability, and shipowners' mutual insurers still do not recognize remote-control-center staff as seafarers. US driverless trucking has no national safety framework, only a growing patchwork of state laws. Until liability is settled, expect insurers and legal teams, not engineers, to set the pace.\n\n- **Scale is running well ahead of the economics.** Aurora lost $270M in the second quarter on $2M of revenue, and every major robotaxi operator is still losing money. Today's service prices are partly subsidized by investors, so budget for the possibility that they rise, or that your provider does not last."
    }
  ],
  "url": "https://www.thestateofplay.ai/domain/autonomous-systems-vehicles",
  "license": "CC BY 4.0",
  "licenseUrl": "https://creativecommons.org/licenses/by/4.0/",
  "generatedAt": "2026-10-01"
}