17. Real Robot Teleopartion

• 17. Real Robot Teleopartion
  • 17.1. Forcefeedback Warning
  • 17.2. User Input
  • 17.3 Known Pitfalls

Please use the demos found in alr_sim.sims.sl.multibot_teleop as guidance.

We have demos for three scenarios:

• demo_teacher.py for kinesthetic teaching of robots. The real robot is put into gravity compensation mode and can be guided by gently pushing the robot arm.

• demo_teachermulti.py for kinesthetic teaching of multiple robots, e.g. in a bimanual setup.

• demo_teleop.py for teleoperation. One robot is designated as primary robot in the aforementioned "human teacher" mode. A second robot is designated as replicant and mirrors the movements of the first robot.

• demo_vtwin.py for a virtual twin setup. In contrast to teleoperation, the replicant robot exists only in simulation. This is a useful mode to itneract with objects in simulation without requiring a CV setup for object tracking.

The virtual twin setup is the most complete for now, showcasing a complete experiment with object interaction, logging capabilities and scene resets.

The code makes heavy use of multibot features to control mupltiple robot movements within one process. Please also refer to 07 Multibot Docu.

17.1. Forcefeedback Warning

Teleoperation supports Forcefeedback. We do not recommend enabling Forcefeedback in a virtual twin setting, as the simulated forces might behave strangely.

17.2. User Input

The demo_vtwin.py shows how you can use a simple CLI text input to control the process, such as starting and stopping logging or resetting the scene.

17.3 Known Pitfalls

The HumanController class has an optional regularize=True argument. When set to true, the controller contains a weak force regularizer to keep the robot in a centered position. This is often helpful to get cleaner movements, but might introduce a bias into your data. The regularization also actively fights against reaching exotic joint configurations.

In such cases, remember to turn off the regularization term.