Terminal Constraints for Walking Pattern Generation

[Saeed-Mansouri]

Dear Dr. Caron,

I have encountered an issue with the walking pattern generation for which I appreciate your help. I don’t understand why we should considerer ZMP and DCM terminal constraints at the end of each predictive horizon? My understanding from these constraints is the bipedal robot will end up in a stationary position at the end of the predictive horizon. If my understanding is correct I think both terminal constraints should only be applied at the end of the walking.

[stephane-caron]

The ZMP and DCM constraints at the end of the predictive horizon are a viability condition that guarantees recursive feasibility of the MPC. Check out Ciocca et al. (2017) for details on this.

In practice, the robot does not reach this stationary state while walking as new steps keep being added and predictive trajectories are recomputed accordingly.