Roadmap¶

This roadmap tracks bvlos-sim from the current deterministic validation engine toward a broader BVLOS simulation platform.

Current Status¶

The current codebase implements 53 tickets across the deterministic, stochastic, regulatory, and documentation tracks, including all core estimator, feasibility, scenario, SITL, stochastic, output-format, and batch capabilities:

estimator hardening, static feasibility checks, and scenario runner (Tickets 001–021)
fidelity v2 trajectory, wind, terrain, resource, link, and Dubins divert (Tickets 030–039)
SITL adapter contract, ArduPilot adapter, telemetry recording, and comparison reports (Tickets 040–043)
stochastic state propagation, twin-state EKF observation model, and closed-loop tracking controller (Tickets 047–049)
real-world data fetch scripts and pre-fetched Alpine demo (Tickets 052–053)
GeoJSON/KML route exports, community vehicle profiles, and --format summary (Tickets 055–057)
deliberately infeasible demo, QGC import, batch estimates, and --format csv (Tickets 059–060)
3D geofence altitude bounds, wind-corrected divert energy, and stochastic spatial infeasibility tracking (Tickets 061, 062, 065)
--format checklist, --format profile, energy reserve sensitivity, and size-battery command (Tickets 072–075)

The test suite currently passes with 1116 tests and 9 skipped live or environment-dependent tests.

bvlos-sim remains an engineering validation tool. It is not a flight-safety system, operational approval tool, or complete BVLOS compliance system.

Implemented Capabilities¶

Estimator:

deterministic mission distance and time estimation
raising and non-raising Python APIs
structured failures for invalid, unsupported, and infeasible cases
partial-result handling
fidelity v1 baseline behavior
fidelity v2 turn arcs and fixed-wing circular loiter

Static feasibility:

deterministic phase-based energy model
deterministic resource-system feasibility for onboard battery, external power, and hybrid power configurations
deterministic communication-link feasibility for mission and scenario link systems
reserve-at-landing output
reserve threshold failures
static GeoJSON geofence checks with optional AMSL altitude bounds
static GeoJSON landing-zone reachability checks

Wind and trajectory fidelity:

ConstantWindProvider
LayeredWindProvider
SpatiotemporalWindProvider with quadrilinear interpolation (offline wind grid)
optional transit sub-segment wind sampling
wind-triangle correction for forward-flight transit
station-keep loiter for hover-capable vehicles

Terrain:

TerrainProvider interface
ConstantElevationProvider
GridTerrainProvider with bilinear interpolation (offline uniform elevation grid)
terrain-referenced altitude resolution from mission asset file
structured diagnostics for missing provider and missing coverage

Scenario runner:

scenario.v1 input schema
deterministic timeline construction
observe, lost-link, wind-change, and landing-zone-unavailable events
assertion outcomes: passed, failed, skipped, unsupported
lost-link policy outcomes for rtl, land, loiter, and divert
computed divert route estimates (DivertRouteEstimate) on divert policy outcomes: geodesic distance, TAS transit time, cruise-power energy, reserve after divert, and feasibility flag
policy_action_eq assertions

Uncertainty modeling:

uncertainty.v1 YAML input schema with normal and uniform distributions
seeded Monte Carlo sampling over wind, cruise speed, cruise power, and battery capacity
run_monte_carlo Python API wrapping the deterministic estimator
sample CLI command for uncertainty execution
uncertainty-report.v1 JSON envelope with summary statistics (mean, std, min, p5, p50, p95, max) and deterministic baseline
Markdown rendering for uncertainty reports

Stochastic propagation:

time-stepped particle propagator with Gaussian process noise
twin true/estimated particle state via EKF predict/update cycle
synthetic GPS and battery-meter sensor noise models (SensorProfile)
proportional cross-track and along-track tracking controller (ControllerProfile)
propagate CLI command and stochastic.v1 input schema
stochastic-envelope.v1 output contract with timeline, estimation_error_timeline, cross_track_timeline, reserve_at_landing_wh, feasibility_rate, and spatial_infeasible_count (geofence/LZ failures counted separately from energy failures)

Interfaces and contracts:

estimate CLI command
scenario CLI command
sample CLI command
propagate CLI command
sitl CLI command for contract-only SITL evidence bundles
canonical estimator JSON envelope (estimator-envelope.v7)
canonical scenario JSON envelope (scenario-report.v2)
canonical uncertainty JSON envelope (uncertainty-report.v1)
canonical stochastic propagation envelope (stochastic-envelope.v1)
canonical SITL evidence bundle (sitl-evidence.v1)
canonical SITL comparison report (sitl-comparison.v1)
ArduPilot SITL telemetry artifact recorder for completed evidence bundles
SITL comparison report JSON and Markdown rendering through compare and adapter APIs
Markdown rendering for estimator, scenario, and uncertainty reports
one-line summary rendering for estimator, scenario, uncertainty, and stochastic reports
GeoJSON and KML route export with per-leg energy-margin colouring, landing-zone reachability markers, and geofence conflict flags
community vehicle profiles in examples/vehicles/community/ for DJI Matrice 300 RTK, Wingtra One Gen II, QS Trinity F90+, Autel EVO Max 4T, and generic survey hexacopter
fetch_wind.py (Open-Meteo archive/forecast), fetch_terrain.py (SRTM via srtm.py), and fetch_landing_zones.py (Overpass API) scripts with pre-fetched Alpine demo in examples/real_world/
golden fixture regression tests

Known Limitations¶

Estimator limitations:

no known path-planning model gaps remaining after Ticket 044

Scenario limitations:

no known scenario path-planning model gaps remaining after Ticket 044

Platform limitations:

ArduPilot SITL connect/upload, telemetry evidence, and comparison reporting are implemented through Tickets 041-043
no PX4 SITL adapter yet; Tickets 045 (launch/upload) and 046 (telemetry/evidence)
no REST API; Ticket 050
no web UI; Ticket 050
no reference inputs for calibration and import; Ticket 054
no NOTAM/live airspace integration; Ticket 058
no live comms, UTM/U-space, Remote ID, or traffic integrations; Tickets 070 and 071
no real-world calibration pipeline; Tickets 080-084

Phase Plan¶

Phase 1: Estimator Hardening¶

Status: implemented.

Delivered:

estimator CLI
canonical result envelope
stable exit codes
schema/versioning policy
golden fixtures
package boundary cleanup
structured error-envelope behavior

Exit criterion: estimator runs from CLI with deterministic versioned output.

Phase 2: Static Feasibility Layer¶

Status: implemented.

Delivered:

deterministic energy feasibility
reserve-at-landing checks
geofence loading and route conflict checks with optional altitude bounds
landing-zone loading and reachability checks
structured diagnostics for feasibility failures

Exit criterion: missions can be accepted or rejected by deterministic static feasibility checks with machine-readable diagnostics.

Phase 3: Scenario Runner and Contingency Policies¶

Status: implemented.

Delivered:

scenario schema
deterministic scenario runner
timeline model
event outcomes
assertion outcomes
scenario JSON and Markdown reports
lost-link policy outcome model
policy action assertions
dynamic wind_change event injection

Exit criterion: repeatable scenario execution and machine-readable assertion outcomes.

Phase 4: Wind and Trajectory Fidelity v2¶

Status: implemented.

Delivered:

layered wind provider
sub-segment wind sampling
fidelity mode selection
turn-arc legs
fixed-wing circular loiter
metadata showing the fidelity mode used

Exit criterion: improved fidelity modes are available without breaking v1 contracts.

Phase 4.5: Environmental Model Extensions¶

Status: implemented.

Delivered:

Ticket 032: terrain-referenced altitude execution
Ticket 033: continuous spatiotemporal wind grid
mission YAML asset integration for terrain and wind-grid files
terrain and wind examples
CLI, envelope, Markdown, and fixture coverage

Exit criterion: terrain and wind-grid inputs can be used through the same mission YAML and estimate command path as existing deterministic features.

Phase 4.6: Resource and Link Feasibility¶

Status: implemented.

Scope:

Ticket 034: resource and link feasibility abstractions — implemented
generalized resource systems for onboard battery, external/tethered or optical-fiber power, hybrid power, and future resource-type extension points
generalized communication link systems for direct, mesh, cellular, satellite, Starlink-class, and hybrid failover architectures
integration with mission YAML, vehicle YAML, scenario YAML, existing feasibility reports, scenario assertions, and later live adapter replay artifacts

Exit criterion: energy/resource and communication-link feasibility can be modeled through shared deterministic abstractions instead of one-off battery-only or lost-link-only fields.

Phase 4.7: Scenario Contingency Model Gaps¶

Status: implemented.

Scope:

Ticket 035: dynamic landing-zone availability — implemented
Ticket 036: computed divert routing — implemented
integration with scenario YAML, mission assets, terrain, wind, geofences, landing zones, resource systems, link systems, and existing scenario reports

Exit criterion: contingency outcomes can use the same configured environment and feasibility features as baseline mission estimation.

Phase 4.8: Uncertainty Modeling¶

Status: implemented.

Scope:

Ticket 037: Monte Carlo uncertainty modeling — implemented
uncertainty.v1 YAML-configured uncertainty inputs
explicit opt-in sample CLI command
uncertainty-report.v1 JSON envelope and Markdown rendering preserving deterministic baseline output

Exit criterion: uncertainty analysis composes with existing mission, vehicle, terrain, wind, geofence, landing-zone, resource, link, energy, and scenario behavior without changing deterministic defaults.

Phase 4.9: Bank-Angle Model and Dubins Path Optimization¶

Status: implemented.

Scope:

Ticket 038: bank-angle model and Dubins path optimization

Delivered:

estimator.math.dubins Dubins path-to-point solver (RS and LS path types)
divert route estimates use Dubins path distance (bank-angle-constrained arc + straight) when entry heading and vehicle.performance.turn_radius_m are available; falls back to straight-line geodesic otherwise
fidelity v2 turn arc confirmed as the exact Dubins solution for a same-position heading change (turn_radius_m * |Δθ|); no change to v2 math
entry heading extracted from the last completed leg's ground_track_deg at the action timeline index

Exit criterion: horizontal path planning accounts for bank-angle constraints and heading continuity across transit, turn, and divert segments without changing fidelity v1 behavior or existing public result field names.

Phase 4.10: Path-Planning Model Gaps¶

Status: implemented.

Scope:

Ticket 039: path-planning model gaps

Delivered:

fidelity v2 transit legs adjacent to TURN_ARC legs subtract the tangent-point offset (turn_radius_m * tan(|Δθ|/2)) from path_distance_m; offsets clamped to zero; total path distance now matches the true Dubins-path length through all waypoints
takeoff and landing-transit legs report path_distance_m = vertical_distance_m (3D slant path for purely vertical legs) in all fidelity modes
DUBINS_DIVERT_PLANAR_APPROXIMATION_LIMIT warning added as an interim diagnostic before Ticket 044 retired the warning emission

Exit criterion: fidelity v2 total path distance equals the sum of offset-adjusted transit legs plus turn arc lengths; takeoff and land legs report correct 3D slant path distance.

Phase 4.11: Geodesic Dubins Divert Path¶

Status: implemented.

Scope:

Ticket 044: geodesic Dubins divert path
replaced the single-point planar Dubins target projection with geodesic-aware target geometry boundary sampling
retired the DUBINS_DIVERT_PLANAR_APPROXIMATION_LIMIT warning emission

Exit criterion: divert distance is accurate within 0.5 % of the true Dubins path length on the WGS-84 ellipsoid for routes up to 500 km; the planar limit warning is no longer emitted for correctly handled distances.

Phase 5: SITL Integration¶

Status: partially implemented.

Prerequisites:

Ticket 034: Resource and Link Feasibility Abstractions — implemented
Ticket 035: Dynamic Landing-Zone Availability — implemented
Ticket 036: Computed Divert Routing — implemented
Ticket 037: Monte Carlo Uncertainty Modeling — implemented

Scope:

Ticket 040: SITL adapter contract and evidence schema — implemented
Ticket 041: ArduPilot SITL launch and mission upload - implemented
Ticket 042: SITL telemetry recorder and evidence bundle - implemented
Ticket 043: SITL scenario comparison report - implemented
Ticket 045: PX4 SITL launch and mission upload
Ticket 046: PX4 SITL telemetry recorder and evidence bundle
adapter-local ArduPilot and MAVLink dependencies that stay outside core estimator/scenario execution
adapter-local PX4 dependencies that stay outside core estimator/scenario execution
comparison against existing deterministic scenario outputs

Exit criterion: a scenario can be executed against supported SITL adapters with recorded evidence and compared against deterministic scenario expectations while preserving the deterministic core execution path.

Phase 4.12: Real-World Data and Developer Experience¶

Status: partially implemented. Done: Tickets 052, 053, 055, 056, and 057. Planned: Tickets 054 and 058.

Scope:

Ticket 052: real-world data fetch scripts — fetch_wind.py (Open-Meteo forecast with --departure-time), fetch_terrain.py (SRTM via elevation package), fetch_landing_zones.py (Overpass API); alpine demo example using pre-fetched real assets
Ticket 053: airspace geofence fetch script — implemented with fetch_geofences.py via OpenAIP API key or keyless Overpass fallback; completes the static real-world asset set started in Ticket 052
Ticket 054: reference inputs for calibration and import design — PX4 ULog flight logs and QGC .plan files committed to reference/ with field-mapping design notes feeding Tickets 060 and 080
Ticket 055: GeoJSON / KML route export — implemented with --format geojson and --format kml on estimate and scenario commands; legs coloured by energy margin; landing zones and geofence polygons as separate feature layers; opens directly in Google Earth and QGC
Ticket 056: community vehicle profiles — implemented with five manufacturer-derived YAML profiles (DJI Matrice 300 RTK, Wingtra One Gen II, Quantum-Systems Trinity F90+, Autel EVO Max 4T, generic survey hexacopter) with provenance links
Ticket 057: summary output format — implemented for estimate and scenario commands; single-line go/no-go digest with reserve %, flight time, contingency policy action, and first failing check; suitable for shell scripts and pre-flight checklists
Ticket 058: NOTAM and live airspace integration — fetch_notams.py via FAA B4UFly API (US) and EUROCONTROL NOTAM service (Europe); active TFRs and temporary restrictions merged with static geofence output for day-of-flight feasibility checks

Exit criterion: a new user can fetch real terrain, wind, and airspace data for any location in under five minutes, run a geographically realistic estimate, and open the output in a map viewer without writing any code; community vehicle profiles allow immediate use without placeholder values; summary output is suitable for shell-script automation.

Phase 6: Product Surfaces¶

Status: partially implemented.

Scope:

REST API around estimator and scenario execution
web map UI for route, phases, warnings, failures, and timeline playback
consistent JSON and Markdown report outputs
optional PDF/report export after report contracts stabilize
shared execution path with estimate and scenario

Exit criterion: users can run missions and scenarios from UI or API with consistent outputs.

Phase 7: Stochastic State Propagation¶

Status: implemented.

Scope:

Ticket 047: stochastic state propagation for path-dependent risk — time-stepped propagator carrying a belief state (position, energy, wind estimate) forward; Gaussian EKF-style representation; per-step p_reserve_violation; stochastic.v1 input schema; propagate CLI; stochastic-envelope.v1 output contract
Ticket 048: closed-loop observation model and twin-state architecture — separate true state (physics) from estimated state (autopilot EKF belief); synthetic GPS, battery-meter, and airspeed sensor models; EKF predict/update cycle; policy decisions made from estimated state; estimation_error_timeline output; SensorProfile added to VehicleProfile
Ticket 049: stochastic closed-loop control — tracking controller model (cross-track/along-track error regulator) connecting EKF estimated state back to true trajectory; path-length excess and energy feedback; ControllerProfile added to VehicleProfile; cross_track_timeline output
Ticket 050: contingency trigger probability derived from the twin-state and cross-track timeline
Ticket 051: SITL telemetry replay to condition the belief state retrospectively and validate the controller model against observed tracks

Exit criterion: the simulator can report the probability of reserve violation at any point mid-flight, not only at landing; path-dependent uncertainty accumulates correctly across legs; estimation error feeds back through the autopilot controller into actual trajectory deviation and reserve consumption; the propagation path composes with all existing wind, terrain, and energy models without changing deterministic defaults.

Phase 8: Import, Export, and Batch Workflows¶

Status: partially implemented. Done: QGroundControl .plan importer (convert) and batch estimates (batch). Planned: Tickets 054 and 058.

Scope:

QGroundControl .plan importer — implemented as bvlos-sim convert
broader mission/action compatibility
batch-run support — implemented as bvlos-sim batch with batch.v1 estimate manifests
performance profiling
report comparison and diff tooling
batch manifests referencing existing mission, vehicle, scenario, terrain, wind, geofence, and landing-zone files

Exit criterion: teams can import external plans, run batches, and compare outputs efficiently.

Phase 9: Operational Integration¶

Status: planned.

Scope:

UTM/U-space integration seams
live comms, Remote ID, and traffic-observation seams
operational intent and conformance check interfaces
schema migration guidance
reproducibility and evidence standards

Exit criterion: outputs are operationally integrable, reproducible, and governance-ready.

Real-World Validation Track¶

This track is separate from platform surface area. It improves model credibility through observed flight data.

Planned work:

flight log ingestion (done, Ticket 080)
trace normalization (done, Ticket 080)
flight phase segmentation (done, Ticket 081)
predicted-vs-observed metrics (done, Ticket 082)
calibration profile fitting (done, Ticket 083)
holdout validation reports

SITL is not a substitute for real-world validation. SITL checks behavioral consistency; real logs calibrate and validate model assumptions.

Ticket Backlog¶

The detailed execution backlog is in tickets/README.md.