Wilson extended XCS with interval based conditions to XCSR to handle real valued inputs. However, the possible actions must always be determined in advance. Yet domains such as robot control require numerical actions, so that neither XCS nor XCSR with their discrete actions can yield high performance. In the work presented here, genetic programming-based representation is used for the first time to compute continuous action in XCSR. This XCSR version has been examined on a simple one-dimensional but non-linear testbed problem – the “frog” problem – and compared with two continuous action based systems, GCS and XCSFCA. The proposed approach has consistently solved the frog problem and outperformed GCS and XCSFCA.