Identical reasoning and implementation as constraining geometry lengths to be positive in #8 . During the learning process, the learned friction coefficients could go negative. If they did, the URDF exporting/importing failed. Here, we constrain friction coefficients to be non-negative, by distinguishing the learned parameters (which can take gradient steps during the learning process) from the reported physical quantities (which are used for exporting URDF files and later for simulation).
Identical reasoning and implementation as constraining geometry lengths to be positive in #8 . During the learning process, the learned friction coefficients could go negative. If they did, the URDF exporting/importing failed. Here, we constrain friction coefficients to be non-negative, by distinguishing the learned parameters (which can take gradient steps during the learning process) from the reported physical quantities (which are used for exporting URDF files and later for simulation).