VDSim product roadmap

Last updated: 2026-10-29 · Tests: canonical validation status

Living roadmap from early PoC through v0.8. Tracks what shipped in main vs what is planned. Releases from v0.7.0 on are published to PyPI (pip install vdsim); the per-release detail is in the changelog. The closed-loop plant (python/vdsim_plant.py, tutorial) is on main. Detail specs: docs/design/*; tire phases in design/TIRE_ROADMAP.md.

How to read

• Shipped — in repo, default or opt-in, covered by ctest where applicable
• Planned — scoped in design docs; not yet implemented
• Opt-in — implemented but not default catalog (baseline frozen)
• Thesis-only — on VDSim-Thesis; not merged to main yet

Velocity (recent): v0.6 L5 MBD → v0.7.0 first PyPI release (generic default vehicle) → v0.8.0 aero elements (per-element aero, air_density, L5 downforce).

---

At a glance

Area	Shipped (high level)	Next
Dynamics L1–L5	L1–L3 planar/14-DOF; Ld4 hard-joint; Ld5 stunt + terrain; Ld5 high-fidelity MBD (design); aero elements + L5 downforce (v0.8.0)	see Next
Tire	MF96 + LuGre default; MF2002 .tir (T1); belt transient (T2); per-axle/per-wheel TireSetup	GUI .tir import; T3 VLOW unified low-speed
Drivetrain	Engine inertia + open-diff; 2D torque map + gearbox + shift policy (v2, opt-in); catalog drivetrain_v2 part	Engine-map workshop (UI); L1 powertrain
Brake / steer	Pluggable modules + deadtime; user-defined modules (C++/Python subclass)	Booster/MDPS physics
Catalog / runtime	--scene=, fleet, FMI; make_direct_control_session, FrictionMapConfig, SimOutput wheel GT	External part packs; VDS1 v4; Fx→τ contract (#1)
Closed-loop plant	VDSimPlant and closed-loop evaluation modes on main; generic default vehicle (v0.7.0)	see Next
GUI	3-tab scene UI, catalog API, workshops, multi-vehicle compare	GUI v2 (framework decision); terrain Play; .tir import; expose user-defined modules (decide: C++ .so plugin / Python path / GUI authoring)
Reinforcement learning	Vectorized RL env: VecSession C++ pool (11.9x, bitwise-equal to serial), gymnasium/SB3 adapter, snapshot/restore, domain randomization (§10)	Training-loop ceiling (policy-bound at 16% of core); controller-in-the-loop snapshot
Validation	ISO 7401/4138/3888 re-baselined + CI-gated (IsoBaseline); canonical suite status	Adams x-check rtol; commercial cross-val

timeline
    title Shipped milestones
    section PoC
        L1-L3 + MF96 + FMI : 2025
        ISO maneuvers + driver L5-L8 : 2025
    section v0.2
        Subsystem modules + deadtime : 2026-06
        GUI redesign + multi-vehicle : 2026-06
        Workshops shell : 2026-06
    section v0.3
        Parts catalog M1-M5 : 2026-06
        Drivetrain inertia : 2026-06
        LuGre tire tuned : 2026-06
    section v0.4
        Ld5 stunt jump loop : 2026-06
        Ld4 multibody M1-M7 : 2026-06
    section v0.5
        Terrain + L5 driving : 2026-06
        Tire T1 MF2002 backend : 2026-06
        Tire T2 belt transient : 2026-06
    section v0.6
        L5 free-3D inertial unsprung MBD : 2026-06
        Hardpoint anti-dive/roll-centre + progressive coil : 2026-06
        Gyroscopic wheel-spin coupling : 2026-06
        Per-wheel tire + direct_control session : 2026-06
    section v0.7
        First PyPI release + generic default vehicle : 2026-10-28
    section v0.8
        Aero elements + air density + L5 downforce : 2026-10-29

Next

Plan, not a commitment: scope and order may change, and nothing here is promised for a date.

Version	Content
0.8.1	L5 tyre longitudinal-force torque reaction fix; L5 sessions from Python; error on an unknown level string
0.8.2	L5 hardpoint attach from Python; steady-speed oscillation on the link path; cg_height meaning on the strut path; rack-ratio key; K&C sweep arguments and outputs
0.9.0	Aero stage 2 (ride-height and pitch slopes, user aero module); per-axle tyres
0.10.0	Driver and cone-course API; rocker and U-shaped anti-roll bars

---

1. Dynamics ladder

Item	Status	Notes
[x] Ld1 bicycle (5 state)	Shipped	bicycle_dynamics.cpp
[x] Ld2 seven-DOF planar	Shipped	Per-wheel Pacejka, Ackermann
[x] Ld3 fourteen-DOF (sprung + unsprung)	Shipped	Ride/heave/roll; L2 planar inner
[x] L1↔L2↔L3 cross-model consistency	Shipped	PlanarMotionMatchesL2, grid sweeps
[x] Low-speed kinematic blend (default tire)	Shipped	LOW_SPEED_HANDLING.md
[x] LuGre path skips blend	Shipped opt-in	Full dynamic EOM at all speeds
[x] Ld4 hardpoint kinematics (attach)	Shipped	Native kin YAML; level=L4
[x] Ld4 multibody M1–M7	Shipped	LD4_MULTIBODY.md; M4 runtime, M7 offline
[x] Ld5 6-DOF 3D body + quaternion	Shipped v0.4 (stunt)	free_3d_dynamics.cpp; terrain → v0.5
[x] 3D contact / airborne / ramp jump	Shipped v0.4	V0.4_SLOPE_JUMP_DYNAMICS.md
[x] Ld5 free-3D inertial unsprung MBD (per-corner world point mass + anisotropic 2-point bushing; energy-consistent on any surface; replaces the B1 lumped strut)	Shipped v0.6	L5_MBD_FREE3D_UNSPRUNG.md
[x] Hardpoint-emergent anti-dive/squat + roll-centre migration (wheel follows the real DAE travel path; tyre Fx/Fy → vertical reaction)	Shipped v0.6	active when a corner DAE is attached
[x] Progressive coil rate + damper (k_coil·(l−l_free)·MR from real spring eye-to-eye; bump/rebound asymmetric)	Shipped v0.6	wheel-rate fallback w/o spring hardpoints
[x] Gyroscopic wheel-spin coupling (Σ I_wheel·ω_spin lateral momentum; yaw↔roll at speed)	Shipped v0.6	Stunt.GyroscopicWheelSpinCouplesYawToRoll
[x] Per-substep contact re-query default in SimSession (exact on curved surfaces)	Shipped v0.6	no-op for non-free_3d
Ld5 → full 6-DOF multibody (superseded designs: L5_6DOF_MULTIBODY, L5_MBD_DYNAMIC_COUPLING, L5_MBD_RIGOROUS)	Superseded by free-3D MBD above	kept for history
[x] Terrain + L5 general driving (heightmap hub contact, hill/cliff, inclined)	Shipped v0.5 (headless/batch/cosim; GUI → v0.5.2)	V0.5_TERRAIN_L5.md
[x] Curved banked track (CurvedGround + banked oval)	Shipped v0.5 (was v0.4 M3)	V0.5_TERRAIN_L5.md M5b
[ ] V2V collision	Planned	V0.2_MULTIVEHICLE.md

---

2. Tire

Today: empirical Pacejka MF96 (default) + library MF2002 evaluator + opt-in LuGre brush.

Future layers: design/TIRE_ROADMAP.md (belt transient, catalog MF, …).

2.1 Shipped

Item	Status	Reference
[x] Pacejka MF96 BCDE long/lat	Shipped	pacejka_mf96.cpp
[x] Combined slip (friction ellipse)	Shipped	Ch.03 §3.5
[x] Aligning moment \(M_z\) + trail falloff	Shipped
[x] Load sensitivity \(\mu_{\mathrm{eff}}(F_z)\)	Shipped
[x] Lateral relaxation length → alpha_dyn_	Shipped	Off when LuGre on
[x] Camber thrust (linear API)	Shipped	Ld4 forward
[x] Full Magic Formula .tir evaluator	Shipped lib	magic_formula_tire.cpp; not default catalog
[x] python/tir_to_yaml.py	Shipped	Workshop wire-up open
[x] LuGre decoupled \(z_{\mathrm{long}}, z_{\mathrm{lat}}\)	Shipped default	V0.2_TIRE_LUGRE.md
[x] tire.kinematic_fallback (LuGre off)	Shipped	ISO / legacy baseline
[x] MF96 as breakaway \(g(\cdot)\)	Shipped
[x] Asymmetric \(g\) floors (no long \(\mu F_z\); α-gated lat)	Shipped	2026-06 tuning
[x] ISO \(F_y\) sign (\(F_y = -F_{\mathrm{raw}}\))	Shipped
[x] Peak-axis combined ellipse (LuGre path)	Shipped	Skip triple circular clip
[x] RK4: \(z\) advanced 4× per substep (\(h/4\))	Shipped
[x] fx_kin = clipped LuGre \(F_x\) for wheel spin	Shipped
[x] Catalog preset tire.lugre_on	Shipped
[x] CLI --lugre / --no-lugre	Shipped
[x] GUI LuGre toggle + fleet_overrides persist	Shipped
[x] Scene lugre_grade_demo.yaml	Shipped
[x] LuGreTire/* integration tests	Shipped	Included in the canonical suite
[x] Theory ch.19 + user docs synced	Shipped
[x] Per-axle TireSetup (front/rear yaml, L1/L2/L3/L5)	Shipped 2026-06-20	PerAxleTire.*; catalog tire_rear slot
[x] Per-wheel TireSetup (FL/FR/RL/RR corner yaml)	Shipped 2026-06-20	PerWheelTire.*; catalog tire_fr/tire_rl/tire_rr

2.2 Planned (tire)

Phase	Item	Status
T1	.tir / MF2002 as catalog tire backend (TireParams.backend)	[x] merged
T1	MF2002 combined slip bypasses host friction ellipse	[x] merged
T1	Tire model selectable per-part via backend: mf96 \| mf2002 \| linear — all three live in create_tire_from_params (mf96 default, mf2002→.tir, linear→create_linear_tire), routed by every dynamics (L1/L2/L5).	[x] infra present; TODO: formalize linear (catalog preset + tests + docs), confirm L1/L5 parity. Requested 2026-06-18
T1	GUI tire import + public sample .tir	[ ] (v0.5.2 GUI)
T1	Chrono Pac02 parity gate (ctest -R ChronoPac02Parity)	[x] pure-long ~2% + pure-lat ~1% gated
T1	LuGre \(g()\) from MF2002 \(F_{x0},F_{y0}\) (optional)	[ ]
T2	Belt transient (belt_relax, σ/|Vx| first-order)	[x] merged
T2	Filtered slip → MF96/MF2002 + LuGre (L2/L3/L5)	[x] merged
T2	bicycle (L1) belt wiring (MF + LuGre)	[x] merged
T3	VLOW-class unified low speed (reduce blend reliance)	[ ]
T4	MF2002 \(G_{x\alpha}, G_{y\kappa}\) on dynamic path	[ ]
T5	Turn-slip, inflation, temperature	[ ]
T6	External tire FMU co-sim; public benchmark report	[ ]

2.3 Belt transient (planned — summary)

Steady MF maps slip → force instantly. Real tires lag because the carcass/belt deflects first. Belt transient adds states (e.g. \(q_y\) or \(\alpha_{\mathrm{eff}}\)) with \(\tau \approx \sigma_y / |V_x|\) before the constitutive law — complementary to LuGre (contact bristle, presliding). See TIRE_ROADMAP.md §3.

---

3. Drivetrain

Item	Status	Reference
[x] Pluggable IDrivetrain module	Shipped	WS1 #159–162
[x] Open / locked / LSD split	Shipped
[x] engine_rotational_inertia reflected (\(I \cdot N_f^2\))	Shipped	V0.2_DRIVETRAIN.md
[x] Open-diff coupled 2x2 axle inertia (open_axle_spin_accel)	Shipped	engine inertia felt under symmetric accel, free under wheel-diff
[x] Legacy mode (I_engine = 0)	Shipped
[x] Catalog drivetrain_v1 part type	Shipped
[x] Engine 2D torque map \(T(\mathrm{rpm},\theta)\) (Drivetrain v2, opt-in)	Shipped	theory ch.22; powertrain: block
[x] Multi-ratio gearbox + RPM coupling + gear-dependent reflected inertia	Shipped	EngineGearbox
[x] Shift policy (manual / auto-RPM / user-defined function, pybind callback)	Shipped	set_shift_policy
[x] Engine speed + gear in output (engine_rpm(), current_gear())	Shipped	+ pybind
[x] Idle-floor + slipping launch clutch	Shipped
[x] Catalog drivetrain_v2 part (materialize powertrain into chassis)	Shipped	drivetrain.sedan_v2 part + vehicle.sedan_powertrain blueprint
[ ] Workshop engine map editor (WS2-6)	Planned	Stub in GUI
[x] ISO baseline unaffected (powertrain opt-in, default flat)	Shipped	IsoBaseline green

---

4. Brake & steering

Item	Status	Reference
[x] IBrakeSystem / ISteeringSystem interfaces	Shipped	V0.2_SUBSYSTEMS.md
[x] Proportional brake + front/rear bias	Shipped
[x] EBD by \(F_z\)	Shipped
[x] Ratio steering + rack travel	Shipped
[x] Actuator deadtime (throttle/brake/steer)	Shipped	WS2-0
[x] Steering rack torque in STATE	Shipped
[x] LuGre column friction (separate module)	Shipped	Theory ch.17
[ ] Booster → line pressure → pad \(\mu\)	Planned
[ ] Brake thermal fade	Planned
[ ] ABS / ESC interface	Planned
[ ] MDPS assist, variable ratio	Planned
[ ] Column compliance model	Planned

---

5. Suspension & anti-roll

Item	Status	Reference
[x] Linear spring/damper per corner (L3)	Shipped
[x] IAntiRollBar per-wheel force (#162)	Shipped
[x] Ld4 hardpoint toe/camber into dynamics	Shipped	Fleet susp_kinematics
[x] Catalog susp_ride, susp_kinematics parts	Shipped
[x] Workshops: suspension tab	Shipped	WS2-1
[x] Progressive coil rate + bump/rebound stops (L5)	Shipped v0.6	MR(z_v) from real spring hardpoints; stops on wheel travel; L5_MBD_FREE3D_UNSPRUNG.md
[x] Genuine per-corner unsprung masses (L5 free-3D)	Shipped v0.6	inertial world particles, not body-relative DOF
[ ] Progressive / bump-stop spring curves (L3 catalog)	Planned	L3 ride model (L5 done)
[ ] Unsprung vs sprung damper split (L3)	Planned	PoC backlog
[ ] KC compliance (bushing) under load — full model	Planned	L5 link is a stiff penalty today
[ ] Dependent axle (twist-beam, solid beam)	Planned	Config stubs only
[x] Ld4 multibody M1–M7	Shipped	LD4_MULTIBODY.md; bushing step off in L4

---

6. Actuator, driver & control ladder

Item	Status	Reference
[x] Lc4 pedal (throttle/brake/steer) primary	Shipped	CARLA-compatible
[x] Lc5–Lc8 cascade (PI, PP, waypoint)	Shipped	Demos + theory ch.07–10
[x] Driver latency + noise	Shipped
[x] First-order actuator lag	Shipped
[ ] LQR / MPC (HPIPM)	Planned	PoC Phase 2
[ ] Per-channel actuator nonlinearities (beyond deadtime)	Planned	references/actuator_nonlinearity.md

---

7. Catalog, scenes & runtime

Item	Status	Reference
[x] configs/catalog/ manifest + parts + blueprints	Shipped	v0.3 M1
[x] Legacy flat vehicles/ / tires/ removed	Shipped	M2
[x] vdsim_realtime --scene= only	Shipped	M3
[x] GUI /api/catalog, /api/scene, simconfig v3	Shipped	M4
[x] fleet[] + part overrides materialization	Shipped
[x] Maneuvers configs/maneuvers/	Shipped
[x] Python Vehicle.preset() / Tire.preset()	Shipped
[x] TireSetup pybind + initialize_setup	Shipped 2026-06-20	per-axle / per-wheel
[x] IVehicleDynamics::tire(wheel)	Shipped 2026-06-20	L1/L2/L5
[x] make_direct_control_session + FrictionMapConfig	Shipped 2026-06-20	CmdL1 direct path; make_friction_ground()
[x] SimOutput wheel GT (μ, μ_peak, α/κ peak, tire_forces_wheel)	Shipped 2026-06-20	SimSession::output()
[x] ladder_lowering.hpp (ISO-frozen L1–L3→L4 calib)	Shipped 2026-06-20	was duplicated in dynamics
[x] Steer deadtime via ActuatorModel (not ad-hoc lag)	Shipped 2026-06-20	vp.steer_deadtime_s on direct path
[x] Plant data-access view plant.vehicle.tire.{model,params} (vehicle→part→physics, read-only)	Shipped 2026-06-18	tyre only
[ ] Extend the data-access view to brake/steering/drivetrain/suspension/ARB parts	TODO	needs const module getters on IVehicleDynamics (only set_*_module exists); tyre is the only part with a live model handle today
[x] VLA Plant API P1: roll angle in obs (obs["roll"], obs["pitch"])	Shipped on main	VDSimPlant observation contract
[ ] VLA Thesis Plant API P1: split-μ friction map (left/right different grip)	Planned	create_split_mu_ground exists in C++; needs Python API surface
[x] VLA Plant API P2: steering actuator lag option (selectable 1st-order LPF)	Shipped on main	VehicleParams.steer_deadtime_s; plant tutorial §7
[ ] VLA Thesis Plant API P2: smooth friction patch transition (tanh vs step μ)	Planned	Reduces numerical jitter at patch boundaries
[x] tools/import_part_pack.py stub	Shipped	M5
[x] VLA plant (VDSimPlant, friction patch, obs contract)	Shipped on main and PyPI since v0.7.0	VDSIM_PLANT_TUTORIAL.md
[x] VLA P1: 1-D + 2-D polygon friction map	Shipped on main	PolygonFrictionGround
[x] VLA P2: smooth patch blend (1 m)	Shipped on main	VDSIM_PLANT_TUTORIAL.md
[x] VLA: mu_peak per wheel (realized friction circle)	Shipped on main	fixes useGT>1 artifact
[ ] VLA #0 Fx→τ C++ contract + drive_split_front (#6)	Planned	spec in VLA_THESIS_PLANT.md; thesis uses CmdL1 shim
[ ] C++ in-process catalog resolver	Planned	PARTS_CATALOG §12
[ ] External catalog zip import (full)	Planned
[ ] VDS1 protocol v4 (vehicle_id field)	Planned	Multi-vehicle I/O
[ ] In-engine V2V collision	Planned

---

8. GUI & workshops

Item	Status	Reference
[x] Full-screen app.html (3D + telemetry)	Shipped	v0.2 redesign
[x] 3-tab setup (vehicle / env / simulator)	Shipped
[x] Multi-vehicle fleet tree + viz	Shipped
[x] Data-comms panel (UDP targets)	Shipped	2026-06 week
[x] Scene save/load (simconfig v3)	Shipped
[x] Workshops window (susp/brake/steer/drivetrain)	Shipped	WS2
[x] Live tire/vehicle edit → plant	Shipped
[x] Catalog part picker in GUI	Shipped	v0.3 M4
[ ] Tire workshop: .tir import button	Planned	WS2-5
[ ] Engine map workshop (curves)	Planned	WS2-6
[ ] Stunt authoring preset panel (ramp / loop / banked)	Planned v0.5 (render-only today)	V0.5_TERRAIN_L5.md M5c

---

9. Co-simulation, FMI & integrations

Item	Status	Reference
[x] VDS1 UDP CMD/STATE (vdsim_realtime)	Shipped	200 Hz class
[x] FMI 2.0 export + round-trip Δ=0	Shipped
[x] Multi-vehicle single process	Shipped
[x] CARLA raycast ABI skeleton	Shipped	PoC ~45%
[ ] CARLA UE5 full sensor bridge	Planned	PoC backlog
[ ] External MF-Tyre / tire FMU co-sim	Planned	Tire T6
[ ] Commercial reference-tool cross-validation	Planned	License + confidential data

---

10. Reinforcement-learning environment

Training surface on top of SimSession; default-path physics unchanged.

Item	Status	Reference
[x] VecSession — N resident sessions on a C++ thread pool, GIL released across the barrier	Shipped	core/src/vec_session.cpp; bitwise ≡ serial stepping, 11.9x on 22 threads, 6.8% barrier overhead
[x] Gymnasium / SB3 adapter (VDSimEnv, VDSimVecEnv, make_sb3_vec_env)	Shipped	python/vdsim_rl/; PPO 100k-step run learns lane-keep + speed-hold (return 87.5 -> 552, episode length 309 -> 967 of 1000)
[x] Six termination conditions judged in C++ with reason codes	Shipped	off-track, rollover, spin-out (beta / yaw-rate), stall, time limit
[x] Zero-copy (N, obs_dim) float32 observation buffer, YAML field set	Shipped	configs/rl/default_env.yaml
[x] Runtime domain randomization: mu, mass, tire stiffness / mu, sensor latency	Shipped	reset() arguments; provider re-creation 0
[x] Per-env seed streams (python + core), bitwise reproducible	Shipped	reset() also re-arms the diagnostic snapshot and the sensor RNG
[x] settle_spawn_on_ground in core, exposed as reset(settle=True)	Shipped	moved out of cosim/; every front end settles a spawn the same way
[x] SessionSnapshot save / restore (actuator, delay lines, sensor RNG, tire relaxation, model lag)	Shipped	picklable; restore is bitwise-identical on L1 / L2 / L3
[x] Substep accuracy gate vs a 0.1 ms baseline	Shipped	1.0 ms 1.22% / 2.5 ms 4.07% / 5.0 ms 6.82% -> training 2.5 ms, evaluation 1.0 ms, 5.0 ms rejected (configs/rl/fast_env.yaml)
[ ] CascadeController and non-identity network snapshotting	Planned	explicitly out of the first cut
[x] Throughput target, re-judged 2026-09-17	Lowered / re-defined	The old "1.0 M ticks/s" was a core-only number with no hardware basis. Measured on the reference box (8P+16E): core 378 k ticks/s @ 22 threads, near-linear on P-cores (1.97x @ 2 th) and ~half-rate on E-cores (11.8x @ 22 th). 1.0 M would need ~31 P-core equivalents -- a hardware statement, not a code one. New target: >=350 k ticks/s on the reference box and >=90% of core throughput retained through the python adapter -- both met (378 k / 93.7%).
[ ] Training-loop throughput (the real ceiling)	Open	End-to-end PPO reaches only 61 k ticks/s = 16.2% of core; the CPU policy forward/backward dominates, and 64 envs is slower than 16 (32.8 k vs 61.3 k) because torch and the sim pool contend for the same cores. Candidates: GPU policy, pinning the sim pool off the torch threads, larger n_steps per update. Measured by tests/rl/bench_throughput_layers.py.

---

11. Validation & credibility

Item	Status	Reference
[x] Analytic benchmarks (bicycle, coast, weight transfer)	Shipped	VALIDATION.md
[x] ISO 7401 step-steer runner	Shipped
[x] ISO 4138 understeer gradient	Shipped
[x] ISO 3888-2 DLC metric	Shipped
[x] ISO 8608 road PSD classes	Shipped
[x] Automated ctest suite; current result	Shipped	2026-06-17
[x] ISO matrix re-baseline (post engine inertia + LuGre) + CI gate	Shipped	IsoBaseline; VALIDATION.md table 2026-06-10
[x] Stunt / L5 validation suite	Shipped	test_stunt, test_l5_strut_validation (loop critical, ballistic jump, camber=L4 DAE, gyro, energy)
[x] KC geometry cross-validation (L5 DAE travel ≡ independent native kinematics, rtol-gated)	Shipped	MultibodyKcXval.DaeTravelMatchesNativeKinematics; rules out engine bugs (internal consistency)
[x] KC analytic anchor (pure-lateral trailing arm ⇒ camber/toe gain = 0, closed form)	Shipped	MultibodyKcAnalytic.PureTrailingArmZeroCamberToe; validates the engine vs math, not just itself
[x] External Chrono KC cross-val (camber/toe vs travel vs Chrono::Vehicle ChSuspensionTestRig, same hardpoints)	Reference CSV + gate green	external/chrono_kc/ + ChronoKcParity (8 % / 0.1° bands; rebound reach ~−14 mm)
[ ] External Adams / physical KC-rig cross-val (absolute, third-party)	Planned	Chrono is an independent MBD but still a simulator; physical/commercial data remains the gap
[ ] Published commercial cross-val (open data)	Planned	Honest gap today

---

12. v0.4 — stunt physics (planned release)

Single tag v0.4.0 per V0.4_PLAN.md. Does not block tire T1/T2.

Milestone	Item	Status
M1	T23 jump C++ + contact is_valid	[x]
M2	RampGround + ramp scene + GUI render	[x]
M3	CurvedGround + banked scenario	→ v0.5 (M5b)
M4	Ld5 core + jump on Ld5	[x]
M5	LoopGround + vertical_loop_demo.yaml	[x]
M6	Loop validation (✓) + theory 20_ld5_stunt	[x]

v0.4.0 tagged. Descoped to v0.5 (now shipped): banked curve (CurvedGround/banked_oval, M5b). Still deferred to v0.5.2 GUI: stunt authoring panel (render-only today).

---

13. Version history (checklist)

PoC → v0.1 (2025)

• L1–L3 dynamics + MF96 tire
• Combined slip, \(M_z\), diff, EBD, aero
• FMI 2.0, Python bindings, ISO runners
• Driver L5–L8, scenarios (step, DLC, skidpad, …)

v0.2 — composable vehicle (2026-06)

• WS1 subsystem modules (brake, steer, drivetrain, susp, ARB)
• WS2 workshops shell + deadtime bridge
• WS3 scene UI redesign + data-comms
• WS4 part registry + fleet assembly
• Multi-vehicle runtime (single process)

v0.3 — parts catalog (2026-06)

• M1 catalog resolver + manifest
• M2 migrate configs, delete legacy trees
• M3 --scene= CLI
• M4 GUI catalog API + simconfig v3
• M5 docs sweep + import stub
• Drivetrain engine_rotational_inertia
• LuGre tire default on + kinematic_fallback part + tests + docs

v0.4 — stunt / Ld5 (2026-06, tagged)

• Stunt / Ld5 / 3D contact (this doc §11)
• Ld4 multibody M1–M7

v0.5.0 — terrain + L5 (2026-06, tagged)

• Terrain + L5 general driving (heightmap hub contact, hill/cliff/inclined, banked)

v0.5.1 — tire layers (2026-06, tagged)

• Tire T1 MF2002 .tir backend + combined-slip ellipse bypass
• Tire T2 belt transient (L2/L3/L5, MF+LuGre, validation, theory ch.21)
• ISO re-baseline (flat, sedan L2 LuGre) + CI gate IsoBaseline

v0.6.0 — L5 high-fidelity MBD (2026-06, tagged)

• Ld5 free-3D inertial unsprung MBD (per-corner world point mass + anisotropic 2-point bushing; energy-consistent on any surface; replaces the B1 lumped strut)
• Hardpoint-emergent anti-dive/squat + roll-centre migration (wheel constrained to the real DAE travel path; tyre Fx/Fy → vertical reaction)
• Progressive coil rate + damper from the real spring eye-to-eye geometry (MR(z_v), bump/rebound asymmetric); wheel-rate fallback without spring hardpoints
• Gyroscopic wheel-spin coupling (yaw↔roll at speed)
• Per-substep contact re-query default in SimSession; contact sampled at the wheel particle (exact on curved surfaces, no false contact leaving a loop/bank)
• Canonical suite status tracked in Validation; design L5_MBD_FREE3D_UNSPRUNG.md

v0.6.1 — core port from thesis (2026-06-20, on main)

• TireSetup per-axle / per-wheel + catalog cosim slots + PerAxleTire / PerWheelTire tests
• make_direct_control_session, FrictionMapConfig, polygon friction canonical path
• IVehicleDynamics::tire(wheel), SimOutput wheel GT fields
• ladder_lowering.hpp; steer deadtime via actuator on direct-control path

v0.6.2 — core quality + tire T3 (2026-06-20, on main)

• SessionKind (Standard / DirectControl) replaces SimConfig.direct_control_path
• KC bushing compliance in L3/L4 hard-joint step (compliance_targets_rad under tire load)
• T3: TireParams.low_speed_mode (kinematic_blend | tire_vlow) + vlow_speed_eps

v0.7.0 — first PyPI release (2026-10-28, tagged)

• Published to PyPI (pip install vdsim); wheels for CPython 3.8–3.12 (Linux x86_64, Windows AMD64)
• Generic default vehicle (generic_sedan + generic_pacejka); non-public vehicles via $VDSIM_PRIVATE_CONFIGS
• Closed-loop evaluation modes, live 3D monitor, RL connection layer, actuator library (see the changelog)

v0.8.0 — aero elements (2026-10-29, tagged)

• aero.elements (per-element downforce and drag at its own point), air_density, aero_drag_height
• Four-key lift now acts on L5; L5 applies aero as a body wrench (theory 26)
• Opt-in aero trace channels (schema 0.5 unchanged)

v0.5.2+ (planned)

• GUI terrain Play (M4) + stunt authoring (M5c) + .tir import — browser-gated
• Tire: bicycle (L1) belt (MF+LuGre) + Chrono Pac02 parity gate — done
• Engine-map workshop UI (drivetrain v2 already in core)
• Brake/steer physics upgrade
• L5 next cut: KC compliance model, gyro axis tilt + spin-up reaction
• GUI v2 (framework + design system — see HANDOFF)
• CARLA full bridge · open benchmark

---

14. External messaging (approved)

Shipped today

Open-core L1–L5 vehicle dynamics with Pacejka MF tire, optional LuGre brush + belt transient layers, opt-in MF2002 .tir backend, per-wheel tire setup and drivetrain v2 (engine map + gearbox), parts catalog, real-time UDP/FMI, and ISO-standard validation. The L5 model is a hardpoint-driven free-3D multibody: anti-dive/squat, roll-centre migration and a progressive coil rate emerge from the suspension geometry, energy-consistent on arbitrary surfaces (banks, loops, jumps).

In active development

Adams/KC cross-validation of hardpoint-emergent geometry; full bushing compliance; workshop importers; GUI terrain play; VLA closed-loop plant (thesis channel, merge to main after client sign-off) — default baselines frozen per phase.

Do not claim (yet)

MF-Tyre product parity, published real-vehicle cross-validation, Adams-validated KC numbers, or production sign-off.

---

15. Related documents

Doc	Role
PLANT_CUSTOMER_FEEDBACK.md	BETA #1 customer feedback (VLA thesis); P1/P2 feature requests
VDSIM_PLANT_TUTORIAL.md	External closed-loop plant user guide
PLANT_DELIVERY_NOTE.md	Thesis-side handoff (setup, API, validation)
design/VLA_THESIS_PLANT.md	Plant specification
design/V0.2_PLAN.md	v0.2 workstreams (historical + status)
design/V0.4_PLAN.md	Stunt / Ld5
design/V0.4_SLOPE_JUMP_DYNAMICS.md	Grade, terrain, T23 jump path
design/L5_MBD_FREE3D_UNSPRUNG.md	L5 free-3D MBD (shipped): unsprung, anti-dive/roll-centre, progressive coil, gyro
design/L5_6DOF_MULTIBODY.md, L5_MBD_DYNAMIC_COUPLING.md, L5_MBD_RIGOROUS.md	Superseded L5 MBD designs (history)
design/TIRE_ROADMAP.md	Tire phases T1–T6 detail
design/V0.2_DRIVETRAIN.md	Engine inertia
design/V0.2_TIRE_LUGRE.md	LuGre shipped spec
design/PARTS_CATALOG.md	Catalog M1–M5
VALIDATION.md	Evidence matrix
HANDOFF.md	Agent handoff / current sprint
design/VLA_THESIS_PLANT.md	VLA closed-loop plant spec (thesis channel)
VDSIM_PLANT_TUTORIAL.md	VLA plant user guide (thesis branch)

---

16. Development priorities (main, 2026-06-20)

Historical snapshot (2026-06-20). The current plan is the Next table.

Ordered for implementation (not doc-only). ISO / default subsystem numbers stay frozen unless an explicit re-baseline is agreed.

P	Track	Task	Why now
P0	Release	Tag v0.6.1 (core port) + v0.6.2 (quality/T3)	v0.6.0 already tagged
P1	Contract	#1 Fx→τ C++ path + force-intent API (VLA_THESIS_PLANT.md §#0)	Blocks clean plant merge; thesis uses CmdL1 shim
P1	Contract	#6 drive_split_front vs brake bias semantics	AWD thesis acceptance; design conflict with brake_bias_front
P2	API	#2 SessionKind generalisation (plant_path / direct control)	[x] 2026-06-20
P2	L5	KC compliance (bushing under load)	[x] 2026-06-20 (L3/L4 step; quasi-static bushing)
P2	Validation	Chrono KC offline build → un-skip ChronoKcParity	[x] 2026-06-20 (reference CSV + gate green)
P4	Tire	T3 VLOW unified low-speed (reduce kinematic blend reliance)	[x] 2026-06-20 (low_speed_mode, vlow_speed_eps)
P3	GUI	v0.5.2: terrain Play or .tir import (pick one)	Browser-gated; no headless path
P3	GUI	Framework decision (React/Svelte + Figma) before more app.js polish	HANDOFF: external review 4/10
—	Thesis	VLA hotfix only on VDSim-Thesis; merge plant PR when client done	vdsim-thesis-beta frozen baseline

Explicitly not now: V2V collision, GUI full rearchitecture without framework choice, MF-Tyre parity claims, ISO number re-baselining.