In AI research, Relational Reinforcement Learning (RRL) is a vastly discussed domain that combines reinforcement learning with relational learning or inductive learning. One of the key challenges of inductive learning through rewards and action is to learn the relations incrementally. In other words, how an agent can closely mimic the human learning process. Where we, humans, start with a very naive belief about a concept and gradually update it over time to a more concrete hypothesis. In this paper, we address this challenge and show that an automatic theory revision component can be developed efficiently that can update the existing hypothesis based on the rewards the agent collects by applying it. We present a symbolic reinforcement learning framework with the automatic theory revision component for incremental learning. This theory revision component would not be possible to build without the help of a goal-directed execution engine of answer set programming (ASP) - s(CASP). The current work has demonstrated a proof of concept about the RL framework and we are still working on it.