Unknown Constraint

Research on unknown constraints focuses on developing methods to infer and learn constraints from data, particularly expert demonstrations, when explicit constraint definitions are unavailable. Current approaches leverage techniques like positive-unlabeled learning, inverse optimization, and Bayesian optimization, often employing neural networks or Gaussian processes to model complex, potentially nonlinear constraints. This work is significant for enabling safe and efficient planning and control in robotics, optimization problems with black-box constraints, and other applications where complete knowledge of constraints is impractical or impossible.