A Dynamic Safety Shield for Safe and Efficient Reinforcement Learning of Navigation Tasks