fr3v2_1 does not load in Isaac Sim 5.1: rigid bodies nested without resetXformStack
I referenced fr3v2_1 (package simready.hf.frankarobotics.simready-assets.fr3v2_1.2026.10.12_00, root simready_usd/fr3v2_1.usda) into a new stage in Isaac Sim 5.1.0, headless, with the PhysX variant Enabled. PhysX builds no articulation. The log starts with:
Rigid Body of (/World/asset/Geometry/base/fr3v2_1_link0/fr3v2_1_link1) missing xformstack reset when child of another enabled rigid body (/World/asset/Geometry/base/fr3v2_1_link0) in hierarchy. Simulation of multiple RigidBodyAPI's in a hierarchy will cause unpredicted results. Please fix the hierarchy or use XformStack reset.
PhysicsUSD: CreateJoint - no bodies defined at body0 and body1, joint prim: /World/asset/Physics/fr3v2_1_joint5
Failed to find articulation at '/World/asset/Geometry/base'
The first error repeats for link2 to link7, the second for every joint. Each link sits inside the previous one and has RigidBodyAPI, and Isaac Sim 5.1 drops a rigid body under another rigid body unless it resets the xform stack. Your benchmark ran on Isaac Sim 6.1, which probably accepts it. Nothing in the package says it needs 6.x.
When I give the seven nested links !resetXformStack! with their world pose, in the session layer and without touching your files, it loads with 15.93 kg and passes PhysX checks: hold, tracking, joint limits, 50 random poses. Tracking needed the floor removed, because at 75 % of joint2's range the arm reaches it from the home pose.
Links as siblings under one prim (the layout of Isaac Sim's own robots) or the reset on each nested link would make it work on 5.1. Stating the minimum Isaac Sim version would also help.
A guess, not a finding: the pose the geometry is authored in, all joints at 0, is outside the limits of joint4 [-176.3, -6.7] and joint6 [25.2, 264.8]. Your PhysicsJointStateAPI home pose is inside them. The state_accuracy failures in your .metadata benchmark are on exactly these two joints. I did not check whether the two are related.
Thanks for the detailed report and for testing the asset on 5.1. Everything you describe checks out.
Isaac Sim 5.1. We reproduced it. We ran the SimReady benchmark on fr3v2_1 with Isaac Sim 5.1.0, and the log shows the same three errors: "missing xformstack reset" for link1 to link7, "CreateJoint - no bodies defined" for every joint, and "Failed to find articulation". The rendering checks pass, but every articulation test stops because PhysX never builds the articulation. On Isaac Sim 6.1, which is what the published benchmark used, the same asset loads and simulates.
We looked at both fixes you suggested:
!resetXformStack!on each nested link. The reset makes a link ignore every parent transform. If you place the robot under a moved or rotated prim, for example two arms in one scene, link0 moves with it while links 1 to 7 stay where they were authored. In your session-layer test the asset sat at the origin, which hides this.- Links as siblings. This works on 5.1, but it means restructuring every Franka asset and dropping the kinematic-chain hierarchy that the SimReady spec recommends (HI.009).
For now we're keeping the hierarchy and stating the requirement instead. The next package version of every Franka asset in this dataset will carry min_isaac_sim_version = "6.0" in SimReady_Metadata, and we'll note it on the dataset card. You're right that nothing said so before.
Joint4 and joint6. Your guess is correct, and it's the cause of the state_accuracy failures. The all-zero pose isn't reachable on a real FR3: the arm collides with itself. That's why joint4 [-176.3°, -6.7°] and joint6 [25.2°, 264.8°] exclude 0. The geometry is authored in the all-zero pose only because that's the URDF convention. Simulation starts from the PhysicsJointStateAPI home pose, which is inside all the limits.
state_accuracy commands every joint to 0 and then ±step. Joint4 and joint6 stop at their limits, and the errors in the benchmark are exactly the distance from 0 to the limit: 6.7° on joint4 and 25.2° on joint6 with PhysX (22.8° with Newton). The limits behave as intended, so we won't move them to pass the test. We'll suggest that the test centre its targets inside each joint's range or at the authored home pose.
The arm reaching the floor at 75% of joint2's range is expected: the arm is fixed to the world, and from the home pose its reach goes below its own base.
Thanks again, this was very useful.