Constraint learning for control tasks with limited duration barrier functions

When deploying autonomous agents in unstructured environments over sustained periods of time, adaptability and robustness oftentimes outweigh optimality as a primary consideration. In other words, safety survivability constraints play key role this paper, we present novel, constraint-learning framework for control tasks built on the idea constraints-driven control. However, since policies that keep dynamical agent within state infinite horizons are not always available, work instead considers can be satisfied some finite time horizon T > 0, which refer to limited-duration safety. Consequently, value function learning used tool help us find safe policies. We show that, applications, existence is actually sufficient long-duration autonomy. This illustrated swarm simulated robots tasked with covering given area, but sporadically need abandon task charge batteries. how battery-charging behavior naturally emerges result constraints. Additionally, using cart-pole simulation environment, policy efficiently transferred from source task, balancing pole, target moving cart one direction without letting pole fall down.