Controlling the speed of the manipulator/robot

[SujayNiranjan]

Hi, thank you for the great work on cuRobo!

A few setup details for context:

• cuRobo installation mode: Python mode
• Python version: 3.10
• Isaac Sim version: Yes (in use)

---

Issue Details

I'm trying to control the speed of the manipulator when executing tasks — specifically using the examples/isaac_sim/simple_stacking.py script.

In the script (around line 137), there's a commented line:

# self.robot_cfg["kinematics"]["cspace"]["max_acceleration"] = 10.0  # controls how fast robot moves

The comment suggests that uncommenting and modifying this should affect how fast the robot moves. However, even after uncommenting and increasing this value, there was no noticeable change in the manipulator's speed.

When I checked the corresponding franka.yml config file, I noticed that max_acceleration is already set to 15.0, which is higher than the value in the script. So I’m wondering:

---

Questions

1. What are the key parameters that actually control the manipulator's speed?
   From my exploration of src/curobo/wrap/reacher/motion_gen.py (lines ~451–465), I saw that modifying velocity_scale and acceleration_scale could have an effect — but the comments warn that doing so may require retuning the cost terms.

2. What is the correct way to adjust the manipulator speed for testing different cycle times?
   I want to understand how to accurately reflect performance changes both in simulation and in comparison with real-world physics. Ideally, I’d like a clear list of parameters to adjust when tuning for speed, so I can track cycle times and behavior changes properly.

@balakumar-s — would really appreciate any clarification or guidance here.

Thanks in advance!

[balakumar-s]

You should only change time_dilation_factor. This will not change the trajectory optimization and will instead re-time the trajectory, so you should see the same success rate for slow and fast trajectories.

[SujayNiranjan]

Thank you for the reply @balakumar-s!

Adjusting time_dilation_factor does sound like the right approach for scaling timing. Just to confirm my understanding: changing this only re-times the trajectory (keeping the path and feasibility the same), so it’s the recommended knob for experimenting with cycle times while maintaining similar success rates, correct?

I also noticed that when I changed interpolation_dt, the execution time improved quite a bit, since it directly affects how the trajectory is discretized into waypoints. From the documentation, I understand it’s defined as time step for generating interpolated trajectory from optimized trajectory.

Right now I’m testing cuRobo with industrial manipulators in Isaac Sim and measuring cycle times for pick-and-place operations at different speeds. Later, I’ll need to set up a digital twin and publish joint angles via ROS from cuMotion to compare against real hardware and validate whether the cycle times match reality.

So my main question is: if the goal is to make these cycle times physically accurate (not just re-timed for testing), what are the key parameters I should be tuning? Should I rely only on time_dilation_factor, or should I also be adjusting limits like max_velocity, max_acceleration, or interpolation_dt to ensure the robot’s simulated performance matches its real-world capabilities?

Thanks again for the clarification and guidance — this really helps me bridge between simulation and real-world testing!