Real-time A* is a variation of A*. Search continues on the basis of choosing paths that have minimum values of f(node) = g(node) + h(node). However, g(node) is the distance of the node from the current node, rather than from the root node.
Hence, the algorithm will backtrack if the cost of doing so plus the estimated cost of solving the problem from the new node is less than the estimated cost of solving the problem from the current node.
Implementing real-time A* means maintaining a hash table of previously visited states with their h(node) values.