Hey, thank you for your nice work.

I’m currently working on a Libero evaluation pipeline with absolute actions.
My goal is to record absolute actions during data collection (for training) and later replay and evaluate them within the Libero environment.
In my current setup, setting robot.controller.use_delta = False enables absolute control mode. However, I observe that the robot moves gradually toward the target pose rather than reaching it precisely in a single step.
To do so, I’d like to clarify the following:

1. Which robosuite version or commit is used in your Libero evaluation setup?
2. What controller type and configuration (e.g., OSC_POSE, WHOLE_BODY_IK, or a custom variant) are you using for absolute actions?
3. How are absolute actions recorded during data collection — are they captured from controller targets, observations/end-effector poses, or joint states?
4. During evaluation, are these absolute actions executed through the controller (via env.step(action)), or are simulator states directly set for deterministic playback?

Thanks in advance!