Imitation Learning With Stability and Safety Guarantees

A method is presented to learn neural network (NN) controllers with stability and safety guarantees through imitation learning (IL). Convex conditions are derived for linear time-invariant systems NN by merging Lyapunov theory local quadratic constraints bound the activation functions in NN. These incorporated IL process, which minimizes loss, maximizes volume of region attraction associated controller simultaneously. An alternating direction multipliers based algorithm proposed solve problem. The illustrated on a vehicle lateral control example.