Repository navigation
OmniSim v9.2.3
Friction and motor fixes found running robot makers' published models (Neobotix MP-400, Clearpath
Jackal, Ekumen's Andino), the second outside fix to the ROS 2 bridge, and an outside Nav2
benchmark. Each fix has a pin test that fails on the old engine, and each behaviour change carries a
value-parsed hatch that restores the old behaviour. The engine and the physics runtime both change,
so use the v9.2.3 installer (doctor must read bundle == source).
⚠️ Behaviour changes — read before upgrading
- A Solid's own low friction is used. MuJoCo combines two surfaces' friction by taking the
larger, soSolid.newtonFrictioncould only ever raise friction: a box atnewtonFriction 0on a
20° ramp stuck exactly like a box at 1. A shape that states its own friction now wins against a
surface that does not; two stated surfaces still take the larger, so only a stated friction LOWER
than the other side changes anything. Hatch:OMNISIM_NEWTON_FRICTION_PRIORITY=0. - URDF
<gazebo reference="LINK"><mu1>/<mu2>reach the link. They were ignored; they now
become the link'snewtonFriction(the smaller of the two, which keeps a caster's free direction
free). Values above 2 import as 2 and are logged, because MuJoCo's friction cone degenerates at
Gazebo's "never slip" 100: Neobotix MP-400 wheels at 100 spun while the body moved 0 m. The
MP-400 as published now drives 0.842 of commanded travel and turns 0.914 of the commanded angle
(was 0.48 and −0.89). Hatch:OMNISIM_URDF_GAZEBO_FRICTION=0. CylinderandCapsulecolliders take the Solid's own friction. Only Box, Sphere and mesh
colliders did, so the wheel friction imported from<gazebo>tags (Husky 1.0, Jackal 0.5) was
inert. Measured against the hatch: a Jackal's open-loop 4 s pivot 337° → 364° and no longer
depends on the world friction (pivot drift 63 mm → 6 mm); its straight line is unchanged
(1.957 → 1.959 m, kinematic 1.96). The four-Huskyhusky_one_passagerun still succeeds with no
robot contacts and finishes in 106.1 s instead of 102.9 s. Hatch:
OMNISIM_NEWTON_ROUND_SOLID_FRICTION=0.- A motor with zero torque is passive. A URDF
<limit effort="0">imports asmaxTorque 0,
which the Newton bridge drove atmaxVelocityfrom the first step: Andino's passive swivel caster
spun at ~112 rad/s before the drive wheels moved, and the robot turned −72° to +154° for a
commanded 90°. An off motor now commands velocity 0 and its joint is free; Andino turns 80.7°
twice and drives 0.995 of its straights. The importer warns oneffort="0", because
ROS-Industrial descriptions use 0 for "unknown" on joints they do drive: give such a joint a real
effort.OMNISIM_NEWTON_AVAILABLE_TORQUE=0still reverts the torque push.
Fixed
- A stated friction of exactly 0 no longer launches bodies. On every collider type, MuJoCo
clamped it to 1e-5, where its friction cone degenerates, and a sliding box left along the contact
normal at ~850 m/s. A stated friction below 0.001 is now applied as 0.001; the 20° ramp slide is
2.269 m in 1.2 s (frictionless closed form 2.27 m). - ROS 2
ClockNodekeeps publishing when the system clock steps back. Its polling timer ran on
the node's default clock, which follows system realtime, so a backward realtime adjustment stalled
/clock: a −1.9 s step gave a 1.92 s gap with no ticks, and Husky Nav2 runs saw/clockand
/odomgaps of about a second. The timer now runs on a steady clock; the published stamps are
still simulator time. Thanks to @LingZhen07 (#24, #25).
Added
- A manual Husky Nav2 static-obstacle benchmark (
tests/benchmarks/husky_nav2_static_obstacle/):
one cold-start trial around a static box with separate navigation (PASS/FAIL) and evidence
(PASS/INVALID) verdicts, a contact positive control, and 300 engine-free tests. It needs Linux or
WSL2 with ROS 2 Jazzy. Thanks to @LingZhen07 (#27).
What's Changed
- fix(ros2): schedule ClockNode polling with a steady clock by @LingZhen07 in #25
- test: add manual Husky Nav2 static-obstacle benchmark by @LingZhen07 in #27
Full Changelog: v9.2.2...v9.2.3