UR10e Differential IK behaves incorrectly while UR10 works in Isaac Lab 2.3.2

[arambarricalvoj]

Hi everyone,

I'm trying to control a UR10e in Isaac Lab 2.3.2 using the DifferentialIKController, and I'm running into a problem that I can't reproduce with the standard UR10.

I'm using the standard Isaac Lab run_diff_ik.py tutorial as a starting point and adapting it to the UR10e.

I launch the example script with:

./isaaclab.sh -p scripts/tutorials/05_controllers/run_diff_ik.py --robot ur10 --num_envs 10

and UR10 behaves correctly.

To test the UR10e using exactly the same example, I first added UR10e_CFG to the imported robot configurations:

from isaaclab_assets import FRANKA_PANDA_HIGH_PD_CFG, UR10_CFG, UR10e_CFG

Then I added ur10e as a supported robot in the scene configuration:

elif args_cli.robot == "ur10e":
    robot = UR10e_CFG.replace(prim_path="{ENV_REGEX_NS}/Robot")

Finally, for the robot-specific SceneEntityCfg, I added the UR10e configuration:

elif args_cli.robot == "ur10e":
    robot_entity_cfg = SceneEntityCfg(
        "robot",
        joint_names=["shoulder_.*", "elbow_.*", "wrist_.*"],
        body_names=["wrist_3_link"],
    )

I can then launch the same tutorial with:

./isaaclab.sh -p scripts/tutorials/05_controllers/run_diff_ik.py --robot ur10e --num_envs 10

The UR10e loads correctly, but its behavior with Differential IK is very different from the UR10. The robot makes large/unexpected movements, can get into self-colliding configurations, and normally fails to reach the target pose. I'm attaching two videos—one for each robot—so you can see the differences in their behavior.

Screencast.From.2026-10-05.13-59-52.mp4

Screencast.From.2026-10-05.14-00-46.mp4

One thing that I find particularly confusing is the end-effector frame. For the standard UR10 example, the tutorial uses body_names=["ee_link"]. For the UR10e, however, I initially expected to use the same ee_link, since when I inspect the robot USD assets directly in Isaac Sim, both the UR10 and UR10e appear to have an ee_link:

However, when I use SceneEntityCfg.resolve() with the official UR10e_CFG, Isaac Lab reports the following available articulation bodies:

['base_link',
 'shoulder_link',
 'upper_arm_link',
 'forearm_link',
 'wrist_1_link',
 'wrist_2_link',
 'wrist_3_link']

and ee_link is not found. Therefore, for the UR10e I had to use body_names=["wrist_3_link"]. This initially seemed strange to me because the UR10e USD itself appears to contain an ee_link when inspected in Isaac Sim. I then noticed that the official Isaac Lab asset configuration uses different USD files for the two robots. In universal_robots.py:

• UR10_CFG uses: Robots/UniversalRobots/UR10/ur10_instanceable.usd
• UR10e_CFG uses: Robots/UniversalRobots/ur10e/ur10e.usd

So although the robots have very similar kinematics, Isaac Lab is actually loading conceptually different USD assets for them.

Besides, in universal_robots.py, the UR10e also has different actuator parameters and starts from a different default joint configuration:

joint_pos={
    "shoulder_pan_joint": 3.141592653589793,
    "shoulder_lift_joint": -1.5707963267948966,
    "elbow_joint": 1.5707963267948966,
    "wrist_1_joint": -1.5707963267948966,
    "wrist_2_joint": -1.5707963267948966,
    "wrist_3_joint": 0.0,
}

The UR10e configuration also explicitly disables self-collisions:

enabled_self_collisions=False

Hoewever, there is an interesting point: the official UR10e RL environments work.

This makes me wonder whether the difference in the USD hierarchy, articulation bodies, or end-effector frame is relevant to the Differential IK behavior. In particular, I am not sure whether wrist_3_link is really the correct frame to use as the IK end-effector for the UR10e, or whether there is another frame/transform that should be used to reproduce the behavior of the UR10 example.

I would therefore appreciate clarification on what the intended end-effector frame for the official UR10e_CFG is, and whether any additional frame transformation is required when using DifferentialIKController.

Most importantly, what would be the recommended way to make Differential IK work correctly with the official UR10e_CFG?

Is there an existing example, configuration, or known fix for using DifferentialIKController with the UR10e in Isaac Lab 2.3.2? If not, what specifically should be changed in the run_diff_ik.py example — for example, the end-effector frame, Jacobian frame, base frame, target pose convention, or initial joint configuration — to correctly control the UR10e?

I would like to keep using the standard DifferentialIKController rather than switching to a different controller, if possible. Any guidance or working example for the UR10e would be greatly appreciated.

[arambarricalvoj]

I've made some progress by modifying universal_robots.py as follows:

UR10e_CFG = ArticulationCfg(
    spawn=sim_utils.UsdFileCfg(
        usd_path=f"{ISAAC_NUCLEUS_DIR}/Robots/UniversalRobots/ur10e/ur10e.usd",
        rigid_props=sim_utils.RigidBodyPropertiesCfg(
            disable_gravity=False,
            max_depenetration_velocity=5.0,
        ),
        articulation_props=sim_utils.ArticulationRootPropertiesCfg(
            enabled_self_collisions=False, solver_position_iteration_count=16, solver_velocity_iteration_count=1
        ),
        activate_contact_sensors=False,
    ),
    init_state=ArticulationCfg.InitialStateCfg(
            joint_pos={
                "shoulder_pan_joint": 0.0,
                "shoulder_lift_joint": -1.712,
                "elbow_joint": 1.712,
                "wrist_1_joint": 0.0,
                "wrist_2_joint": 0.0,
                "wrist_3_joint": 0.0,
            },
    ),
    actuators={
        # 'shoulder_pan_joint', 'shoulder_lift_joint', 'elbow_joint', 'wrist_1_joint', 'wrist_2_joint', 'wrist_3_joint'
        "shoulder": ImplicitActuatorCfg(
            joint_names_expr=["shoulder_.*"],
            stiffness=1320.0,
            damping=72.6636085,
            friction=0.0,
            armature=0.0,
        ),
        "elbow": ImplicitActuatorCfg(
            joint_names_expr=["elbow_joint"],
            stiffness=600.0,
            damping=34.64101615,
            friction=0.0,
            armature=0.0,
        ),
        "wrist": ImplicitActuatorCfg(
            joint_names_expr=["wrist_.*"],
            stiffness=216.0,
            damping=29.39387691,
            friction=0.0,
            armature=0.0,
        ),
    },
)

However, during control, the effector appears to remain at a certain offset from the target:

Screencast.From.2026-10-06.10-03-35.mp4

Remember that the end-effector link is still set to wrist_3_link.

[arambarricalvoj]

I found the cause of the issue, so I wanted to document the solution and close this discussion.

The problem was related to the structure of the UR10e USD asset. I created a flattened copy of the ur10e.usd (https://omniverse-content-production.s3-us-west-2.amazonaws.com/Assets/Isaac/5.1/Isaac/Robots/UniversalRobots/ur10e/ur10e.usd) and reorganized its joints/links to match the structure used by the working UR10 asset. In particular:

1. Saved the UR10e USD as a flattened USD.
2. Reorganized the joint/link structure to match the UR10 USD hierarchy.
3. Added/fixed the ee_joint under wrist_3_link.
4. Enabled the ee_joint.
5. Made sure that ee_link was properly included as a rigid body in the articulation.

After these changes, ee_link was correctly detected by Isaac Lab as an articulation body, so I could use:

robot_entity_cfg = SceneEntityCfg(
    "robot",
    joint_names=[".*"],
    body_names=["ee_link"],
)

instead of using wrist_3_link as the end-effector.

There was also a second issue with the initial joint configuration. The original init_state provided for the UR10e caused the Differential IK Controller to behave very poorly and the robot could become unstable/self-collide. I therefore used the same initial joint configuration as the working UR10 configuration:

joint_pos={
    "shoulder_pan_joint": 0.0,
    "shoulder_lift_joint": -1.712,
    "elbow_joint": 1.712,
    "wrist_1_joint": 0.0,
    "wrist_2_joint": 0.0,
    "wrist_3_joint": 0.0,
}

With the modified USD and this initial configuration, the standard run_diff_ik.py example works correctly with the UR10e and ee_link can be used as the end-effector.

One additional observation: DifferentialIKController does not perform gravity compensation. Therefore, when using disable_gravity=False in universal_robots.py, the UR10e shows a small steady-state offset in Z due to gravity and the dynamics of the implicit actuators. This is separate from the original USD/articulation problem. Setting disable_gravity=True removes this offset.

So, in my case, the original problem was caused by the USD/articulation structure, while the remaining small Z offset with gravity enabled is a consequence of using DifferentialIKController without gravity compensation.

Thanks!