Closed zhonghai1995 closed 2 years ago
In that case \mu
will get set to 0
. See here.
https://github.com/HJReachability/ilqgames/blob/d8aad9d2c71b1abdc2dc1f5dce74595ebf4bd928/include/ilqgames/constraint/constraint.h#L112
You're probably right that it could be implemented more efficiently though. Feel free to submit a PR!
Thanks for the quick reply!
Hello Professor Fridovich-Keil, I have a question about the class proximity constraint. In the implementation of the Quadraticize function of the class proximity constraints, it calls the ModifyDerivatives function. In the ModifyDerivatives function, it calculates the derivatives of lambda g(x) + mu g(x) g(x) / 2.
However, since the proximity constraint is an inequality constraint, then when the constraint value is less than 0 (the constraint is satified), should we still add the derivate of mu g(x) g(x) / 2? Because I think for inequality constraints, the augmented lagrangian is L = lambda g(x) + (mu/2)max(0,g(x))^2?