Tag
DiffSlack proposes a differentiable projection layer that enforces nonlinear inequality constraints in neural networks by reformulating them as equalities with learnable slack variables, achieving improved feasibility and planning success in vehicle path planning.