A sliding block, four simulation stacks

Preliminary diagnostic comparison. Analytic 4 cm cube, 100 g, initial speed 1 m/s, nominal friction 0.4, gravity 9.81 m/s², timestep 2 ms. No robot, controller, damping force or imported mesh.

One sliding block. Different contact solvers.PhysX CPU / Isaac SimNewton / MuJoCo-WarpMuJoCo CPU · slideGenesisTravel along x (mm)136.699.9163.2426.56-10.120.00 s0.20 s0.40 s0.60 sForward velocity (m/s)1.080.78990.49990.2098-0.080280.00 s0.20 s0.40 s0.60 sNet contact force Fx (N)0.3118-0.8149-1.942-3.068-4.1950.00 s0.20 s0.40 s0.60 sNet contact force Fz (N)10.497.6714.8552.039-0.77680.00 s0.20 s0.40 s0.60 sCOM height (mm)21.2320.8420.4520.0619.670.00 s0.20 s0.40 s0.60 sAccumulated horizontal contact impulse (N·s)0.008002-0.02101-0.05001-0.07902-0.1080.00 s0.20 s0.40 s0.60 sMeasured traces · common SI scene · no smoothing · native force semantics differ · not an accuracy ranking
StackTravel (full recording)Stays below 1 mm/s (s)Maximum per-step momentum residual
PhysX CPU / Isaac Sim126.422 mm0.2566.23e-08 N·s
Newton / MuJoCo-Warp126.472 mm0.2661.04e-06 N·s
MuJoCo CPU · slide126.471 mm0.26600000000000025.77e-14 N·s
Genesis126.447 mm0.268.86e-14 N·s

The ideal non-rotating Coulomb slider stops after 127.42 mm and 0.25484 s. This is a reference calculation, not ground truth for a compliant, rotating 3D cube. Equal friction coefficients do not equate contact models.

Normal and friction forces are summed on the cube in world coordinates. PhysX uses step impulses divided by dt; other stacks expose native solver force samples. Do not compare individual contact peaks without considering contact count, compliance and force sampling. The audit checks mΔv = ΣFΔt; it verifies recording consistency, not physical accuracy.

Download vector figure · Raw metrics