[Question] spatial tendons

[Xyunze]

Question

I am a new developer working with Isaac Sim and aiming to train a reinforcement learning (RL) policy using a self-designed robot that includes both angular and linear motors.

I discovered that while angular motors are well-documented, there are no clear examples or guidance on using linear motors with closed-loop control. In my search for alternatives, I came across the concept of spatial tendons as a potential substitute for linear actuators.

However, I have not found any examples or documentation detailing how to configure the articulationCfg for spatial tendons, similar to how it is done for the Unitree H1 in RL locomotion examples.

I would greatly appreciate any guidance on whether it is possible to simulate and train a policy using either linear motors or spatial tendons in Isaac Sim, and how to properly configure them in the RL pipeline.

Thank you!

[RandomOakForest]

Thank you for posting this question. I'll move the post to our Discussions section for the team to follow up.

[RandomOakForest]

Following up, you are correct that Isaac Sim and Isaac Lab provide extensive support for angular (revolute) motors, but working with linear actuators and spatial tendons is less straightforward. Here is a breakdown of the current capabilities and best practices.

1. Simulating Linear Motors in Isaac Sim & Isaac Lab

Linear motors are typically implemented as prismatic joints in USD robotics assets. These can be actuated using either:

• Implicit (Physics-Driven) Actuators: Using the internal PD controller built into PhysX/Isaac Sim. Configure these with parameters like stiffness and damping for closed-loop behavior directly in the actuator config or in USD.¹²
• Explicit Actuators: Define your own dynamic model (e.g., ideal PD or DC-motor) for more realistic or specialized behaviors.²

A typical setup in your articulationCfg for a prismatic (linear) joint would look similar to this (using implicit actuators):

from isaaclab.actuators import ImplicitActuatorCfg

linear_actuator = ImplicitActuatorCfg(
    joint_names_expr=["my_prismatic_joint"],  # name of your linear joint
    stiffness=100.0,                          # spring constant (N/m)
    damping=5.0,                              # damping factor
    effort_limit_sim=400.0,                   # Newtons (force)
)

Add this actuator config to the actuators dictionary in your ArticulationCfg.³²

Key Notes:

• Both prismatic and revolute joints are supported by implicit actuators—no distinction in high-level configuration. Actual closed-loop control is achieved via tuning stiffness and damping for your task.
• Limits (like max extension, velocity) are specified in USD and optionally in the actuator config via velocity_limit_sim, effort_limit_sim.

2. Spatial Tendons as Substitutes for Linear Actuators

Spatial tendons offer a way to simulate mechanisms that are not easily captured by simple prismatic joints—for example, cables, belts, artificial muscles, or hydraulic devices that apply forces between arbitrary points of your robot.

Spatial Tendon Concept:

• Imposes a distance constraint between two or more attachments on your robotic links using spring-damper physics.
• Highly flexible—can pull, branch, and provide nonlinear force transmission.⁴

Configuration Approach:

• Spatial tendons are authored in your USD assets (in Omniverse/Isaac Sim’s stage GUI or programmatically).
• There is currently no direct support in Isaac Lab for configuring spatial tendons via articulationCfg.
• Tendon parameters (stiffness, damping, rest length, etc.) must be set using the Omniverse GUI or USD manipulation tools.⁴
• You can expose tendon rest length as an adjustable parameter and modulate it within the RL pipeline by changing the target rest length, but you’ll need to use lower-level USD or Omniverse Python APIs to update this during simulation.

Summary Table

Mechanism	How to Configure	RL Integration	Documentation References
Prismatic Joint	ArticulationCfg actuator	Direct (action=pos/vel/force command)	123
Spatial Tendon	USD Asset (not ArticulationCfg)	Indirect (change USD attribute at runtime; more advanced)	4

3. RL Policy Training

For both methods:

• Observations and actions can be configured in Isaac Lab as with other joints.
• Closed-loop behavior depends on proper parameterization of stiffness, damping, and velocity/force limits.²³
• For spatial tendons, RL can control the rest length or target force of the tendon by updating the relevant USD attributes from Python, but this is an advanced workflow and may require custom wrappers.

4. Additional Expert Tips

• Study the Cartpole and mobile base examples in Isaac Lab. The Cartpole example uses a prismatic actuator; replicate the actuator configuration for your linear joints.³
• Debug and tune your joints and tendons interactively in Omniverse Isaac Sim before integrating with Isaac Lab RL pipeline.
• For hybrid robots (angular + linear), simply include both revolute and prismatic joints in your robot’s USD and configure actuators as needed in the same ArticulationCfg.²³

5. Key Documentation

• Actuators in Isaac Lab: Covers configuration for both angular and linear actuators, solver- and model-level limits, and typical actuator models.²
• Writing ArticulationCfg: Detailed guide with configuration blocks and physical/real-world mapping.³
• Spatial Tendons: Omniverse core documentation on spatial tendons, authoring, and property configuration (see Snippets > Spatial Tendon Actuation in Isaac Sim for a working demo).⁴

Bottom Line

• Linear actuators (prismatic joints): Fully supported via Isaac Lab’s standard actuator config; use as you would with angular motors, just referencing your prismatic joints.
• Spatial tendons: Supported in USD/Isaac Sim and can be used for RL, but require tendon-specific setup at the asset (USD) level and custom Python scripting for RL control. There is no built-in Isaac Lab config API for tendons as of July 2025.
• For both, closed-loop behavior is set with PD gains and solver limits.

For implementation, always refer to the specific Isaac Lab and Omniverse documentation for your Isaac Sim version. If your workflow requires on-policy changes to tendon parameters, prepare to extend the standard Isaac Lab stack with custom USD/Omniverse scripting.¹⁴³²

Footnotes

1. https://isaac-sim.github.io/IsaacLab/main/source/overview/core-concepts/actuators.html ↩ ↩² ↩³

2. https://isaac-sim.github.io/IsaacLab/main/source/api/lab/isaaclab.actuators.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

3. https://isaac-sim.github.io/IsaacLab/main/source/how-to/write_articulation_cfg.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

4. https://docs.omniverse.nvidia.com/extensions/latest/ext_physics/articulations.html ↩ ↩² ↩³ ↩⁴ ↩⁵