Implement the COBYLA algorithm, a gradient-free method for constrained nonlinear optimization. It models both the objective and constraint functions with linear approximations and uses a trust-region approach. Suitable for problems with nonlinear inequality constraints and no gradient information.

I could not find the original paper online (at least without a paywall). This paper here explains the collection of Powell's derivative-free algorithms in chapter 3: https://arxiv.org/pdf/2302.13246