Skip to content

NVIDIA PhysX, a laden 44-tonne lorry

Lorry tipping point

Before a simulated lorry can say anything about real roads, it has to tip where physics says it should. We built a fully laden 44-tonne articulated lorry in NVIDIA PhysX, NVIDIA's physics engine, and put it through two checks with the pass marks written down before anything ran: a tilt table, and steady turns on flat circles. Passing was the condition for using it on real road layouts and comparing it with police records of lorries that overturned (the crash-record test).

Simulated for one laden lorry on level test surfaces. Real lorries vary with load, tyres, suspension and driver.

The physics check Not passed

The physics check failed again in round 2 (25 m circle), so the crash-record test was not run.

Round 1: The physics check failed (100 m circle, 200 m circle), so the crash-record test was not run.

So the simulated lorry was not used on any real road layout, and no police crash record was compared with it. The rules fixed before round 2 ran say a second failure is final.

SimulatedNVIDIA PhysX 5.9.1Modelledcircle targets, from the threshold

The film

44 seconds, silent. Each lorry shot is drawn from NVIDIA PhysX's own output for round 2: where the lorry is, how far it leans and which wheels carry load, frame by frame. A wheel shows red while it carries no load.

Simulatedround 2, NVIDIA PhysX

Round 2 Not passed

Round 1's fault was in our model, so we were allowed one more round, with every changed value taken from a published source and the rules for it committed before it ran. What changed:

  • The tractor's weight and its split between the axles, from a manufacturer's data sheet for a 6x2 tractor, and its centre-of-gravity height from a US Federal Highway Administration report. Round 1 used estimates.
  • Tyre cornering stiffness for each axle, from measurements on a three-axle truck (Winkler and Aurell, UMTRI with Volvo), in place of one estimated value for every tyre.
  • Cornering load moves to the outside tyres partly through each axle's roll centre, using the roll-centre heights round 1 already cited.
  • Steering by a published driver model for articulated lorries (Pauwelussen): it looks 1.8 seconds ahead and reacts with a 0.1 second lag.

Several values remain marked as estimates in the round 2 vehicle sheet, among them how tyre grip changes with load and the shortest look-ahead at low speed (8 m).

With these values the tilt-table threshold fell from 0.34 g to 0.26 g, which lowers every circle's target speed by about 12%.

Tilt table Passed

Tilted sideways in steps of a tenth of a degree, the trailer's wheels left the platform at 14.6 degrees: a static rollover threshold of 0.26 g. The published range for a fully loaded tractor and semitrailer is 0.25 to 0.50 g.

The trailer's wheels were the first to lift; the same report describes real tractor-semitrailers starting to roll at the tractor's drive wheels. Give in the tyres and springs brings the threshold to 59.7% of what a rigid trailer would manage (0.44 g); the report says this give can take it to "as little as 60 percent".

Platform 14.6 degreesTrailer body 20.8 degreesUpper wheels lift

Flat circles Not passed

On the 50 m, 100 m and 200 m circles the lorry tipped within 10% of the speed physics gives for its threshold. Outside the 10% band: on the 25 m circle it left its line 10.5% below that speed.

Simulatedround 2, NVIDIA PhysXModelledtargets, from the threshold

Simulated speed as a share of the physics target (the dashed line is 1, on target; the pass band is 0.9 to 1.1)

25 m circle, outside the 10% band16.0 mph of 17.9 mph
50 m circle24.0 mph of 25.3 mph
100 m circle34.0 mph of 35.8 mph
200 m circle48.0 mph of 50.6 mph
CircleSimulatedTargetDifference
25 mLeft its line16.0 mph17.9 mph10.5% slowerOutside 10%
50 mTrailer wheel lifted24.0 mph25.3 mph5.1% slowerWithin 10%
100 mTrailer wheel lifted34.0 mph35.8 mph4.9% slowerWithin 10%
200 mTrailer wheel lifted48.0 mph50.6 mph5.1% slowerWithin 10%

Target: the square root of (g x radius x 0.2605), the steady speed at which a lorry with this round's tilt-table threshold would tip on a level circle. A run ends when a trailer wheel on the inside of the turn is off the ground for 0.2 s, or the front axle is 1.5 m off its path.

Why round 2 failed

On the 50, 100 and 200 m circles the round 2 lorry tipped within about 5% of the physics target. On the 25 m circle it did not tip: close to its roll limit it needed more steering than the driver model gives, ran wide, and its front axle reached 1.5 m off the line at 16 mph, 10.5% below the target. That is just outside the 10% bar, so the check failed again, and under the rules fixed before round 2 ran there is no third round.

Round 1 Not passed

Tilt table Passed

Tilted sideways in steps of a tenth of a degree, the trailer's wheels left the platform at 18.7 degrees: a static rollover threshold of 0.34 g. The published range for a fully loaded tractor and semitrailer is 0.25 to 0.50 g.

The tractor's drive wheels lifted first, at 11.7 degrees, which is how the same report says real tractor-semitrailers begin to roll. Give in the tyres and springs brings the threshold to 77.6% of what a rigid trailer would manage (0.44 g); the report says this give can take it to "as little as 60 percent".

Platform 18.7 degreesTrailer body 23.6 degreesUpper wheels lift

Flat circles Not passed

On the 25 m and 50 m circles the lorry tipped within 10% of the speed physics gives for its threshold. Outside the 10% band: on the 100 m circle it tipped 25.2% below that speed and on the 200 m circle it left its line 35.0% below that speed.

Simulatedround 1, NVIDIA PhysXModelledtargets, from the threshold

Simulated speed as a share of the physics target (the dashed line is 1, on target; the pass band is 0.9 to 1.1)

25 m circle20.0 mph of 20.4 mph
50 m circle26.0 mph of 28.8 mph
100 m circle, outside the 10% band30.5 mph of 40.8 mph
200 m circle, outside the 10% band37.5 mph of 57.6 mph
CircleSimulatedTargetDifference
25 mTrailer wheel lifted20.0 mph20.4 mph1.9% slowerWithin 10%
50 mTrailer wheel lifted26.0 mph28.8 mph9.8% slowerWithin 10%
100 mTrailer wheel lifted30.5 mph40.8 mph25.2% slowerOutside 10%
200 mLeft its line37.5 mph57.6 mph35.0% slowerOutside 10%

Target: the square root of (g x radius x 0.3385), the steady speed at which a lorry with this round's tilt-table threshold would tip on a level circle. A run ends when a trailer wheel on the inside of the turn is off the ground for 0.2 s, or the front axle is 1.5 m off its path.

Why round 1 failed

Our reading of the runs: above about 35 mph the simulated tractor turns tighter than it is steered, and the steering correction then swings from side to side. On the 100 m circle those swings, not the steady turn, lift the trailer; on the 200 m circle the lorry leaves its line before it tips. Published studies describe a loaded lorry's tractor as stable at these speeds and turning forces, so we read this as a fault in our model rather than how real lorries behave. The tyre cornering values and the steering look-ahead were engineering estimates, marked as such in the vehicle sheet before the check ran, and they are the likely cause. On the tight circles, where speeds stay low, the same lorry tipped where physics says it should.

The lorry

Weight
44,000 kg on six axles, the GB limit for an articulated lorry
Round 2
Tractor 8,465 kg empty, from a manufacturer's data sheet; load 29.0 tonnes
Tractor
Three axles (6x2, drive axle on twin tyres), from a manufacturer's published data sheet
Semitrailer
13.6 m curtainsider on three axles, from a manufacturer's published data sheet
Load
Spread evenly through the whole trailer, the worst case for rollover in the cited report (29.5 tonnes in round 1)
Centre of gravity
Semitrailer 2.34 m above the road; whole lorry 2.08 m in round 2, 2.10 m in round 1
Track
2.04 m between wheel centres

Method

Engine
NVIDIA PhysX SDK 5.9.1 (release 110.3-omni-and-physx-5.9.1, commit 4f5f477), Vehicle2, on a CPU. Tractor and trailer are coupled at the fifth wheel by a joint, as in PhysX's own truck example.
Tipping
Any trailer wheel on the inside of the turn off the ground for 0.2 s without a break, or the front axle more than 1.5 m off its path.
Speed
Held constant. The tipping speed is found by halving the interval between 5 and 70 mph down to 0.5 mph. Steering by pure pursuit in round 1 and by the driver model in round 2.
Pass marks
Tilt table: a threshold from 0.25 to 0.50 g. Flat circles of 25, 50, 100 and 200 m: the tipping speed within 10% of the square root of (g x radius x threshold) on every circle. Both must pass. The same in both rounds.
Round 1 rules
Written and committed on 10 Nov 2026, before any run. SHA-256 85d1e2995b567da83e00cecc814ae69ffdc75394703283c303dd94664931c4cc
Round 2 rules
Committed the same day, after round 1's result and before any round 2 run, with an amendment also made before it ran. SHA-256 828a5283a3833ddf9f676c63c8832c41953e5ae9f5fa150437174ab598df9f34
Vehicle sheets
Every value cited or marked as an estimate, committed before each check ran (round 1 commit 67a0d2931, round 2 commit 0b5955e0c).
Round 1 runs
Three runs. The first showed two faults in our code: the trailer coupling was solved once per time step rather than at every sub-step (so the tyres carried 42.6 tonnes, not 44), and on the tilt table the lorry slid down the platform. The second run failed on the wider circles (17.8% and 29.8% slow at 100 m and 200 m), and checking it found a third fault: our script gave the tyres 1.5 times the cornering stiffness written in the vehicle sheet. The rule allows two reruns after code fixes. The third run, with that corrected, is the round 1 result shown here. No other vehicle value was changed.
Round 2 run
One run. Before the vehicle sheet was committed, the only runs allowed were slow path-following tests at 10 mph or less on made-up roads, to find code faults; no tilt or circle was run. No fault was found after the run, so it was not repeated.

Sources

  • Physics: NVIDIA PhysX SDK 5.9.1, Apache License 2.0
  • Rollover range: Winkler and Ervin, Rollover of Heavy Commercial Vehicles, UMTRI, 1999
  • Round 2 tyres: Winkler and Aurell, Steady-State Turning of Multiaxle Trucks, UMTRI and Volvo
  • Round 2 driver model: Pauwelussen, Closed Loop Performance of Heavy Goods Vehicles, HVTT
  • Round 2 tractor centre of gravity: US FHWA report FHWA-RD-91-012
  • Weights: The Road Vehicles (Authorised Weight) Regulations 1998, as amended, legislation.gov.uk
  • Tractor and semitrailer: manufacturers' published data sheets

MapMap is a member of the NVIDIA Inception program. This is our own test of NVIDIA PhysX, software available to developers; NVIDIA has not reviewed or approved it. NVIDIA and NVIDIA PhysX are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.

Labels: the lorry's tipping speeds and threshold are SIMULATED in NVIDIA PhysX; the circle targets are MODELLED from that threshold; the published range is from the cited report.

More MapMap demos on NVIDIA technology