CoM And Centroidal Support Example
Overview for examples/04_com_constraint_example.py.
What It Demonstrates
configure_com_constraint()with a 2D support polygon- opt-in horizontal
CentroidalMomentumTaskdamping - commanded-velocity
configure_capture_point_constraint()enforcement - explicit-state
configure_velocity_zmp_constraint()enforcement throughPositionStepOptions.current_joint_velocity - Safety margin shrinking of the active polygon
- Optional proximity activation near polygon boundary
- Visual feedback for CoM, capture point, and finite-difference ZMP
API Walkthrough
The example keeps the same registered-task IK path and adds a hard CoM support-polygon constraint:
| Step | API calls | Purpose |
|---|---|---|
| Register IK tasks | add_frame_task(...), add_posture_task(...), add_centroidal_momentum_task(...) |
Track the end-effector, bias posture, and optionally damp horizontal momentum. |
| Configure support polygon | configure_com_constraint(...) |
Keep the 2D CoM projection inside the active polygon. |
| Configure dynamic support | configure_capture_point_constraint(...), configure_velocity_zmp_constraint(...) |
Keep commanded capture point and finite-difference ZMP inside the same active polygon. |
| Tune relaxation | frame_task.solve_mode, allow_min_error_fallback, posture_task.solve_mode |
Explore strict scaling versus minimum-error fallback near constraints. |
| Supply state | step_opts.current_joint_velocity = dq_current |
Provide the measured or caller-integrated velocity required by physical velocity ZMP. |
| Solve update | solve_position_step(q_current, target_pose, "ee_task", step_opts) |
Apply one IK update with all enabled centroidal constraints. |
| Inspect diagnostics | result.task_modes_effective, result.task_used_fallback, result.task_scales |
Display effective mode, fallback use, and scale while the demo runs. |
Use examples/04_com_constraint_example.py for the exact slider values,
visualization markers, and support-polygon setup.
Run
Install and copy the example bundle once using the Installation Guide. Then run:
For repository development, use Pixi:
The example defaults to the Panda preset; pass --robot <key> to use another
configured model. It uses ViserUrdf by default; use --visualizer pinocchio
to select the alternate viewer.
The capture-point and ZMP disks show accepted-command diagnostics in the support frame. This remains a fixed-base velocity IK example; successful ZMP rows do not certify floating-base contact forces or friction feasibility.