While testing the integration of the MPC with MANN, we realized that the neural network was generating contacts whose activation and deactivation times weren't divisible by the MPC sampling time. This caused an issue when generating the internal vector of the MPC discretization. To overcome this limitation, we decided to implement a method in the ContactList to resample the contacts so that the activation and deactivation times are divisible with respect to a given sampling time.
Specifically, the activation time is decreased to ensure divisibility, while the deactivation time is increased. This way, we increase the overall contact duration.
Last but not least, the MPC checks that the contact timings are correctly sampled.
While testing the integration of the MPC with MANN, we realized that the neural network was generating contacts whose activation and deactivation times weren't divisible by the MPC sampling time. This caused an issue when generating the internal vector of the MPC discretization. To overcome this limitation, we decided to implement a method in the ContactList to resample the contacts so that the activation and deactivation times are divisible with respect to a given sampling time.
Specifically, the activation time is decreased to ensure divisibility, while the deactivation time is increased. This way, we increase the overall contact duration.
Last but not least, the MPC checks that the contact timings are correctly sampled.