Legged robots such as hexapod robots are capable of navigating in rough and unstructured terrain. When the terrain model is either known a priori or is observed by on-board sensors, motion planners can be used to give desired motion and stability for the robot. However, unexpected leg disturbances could occur due to inaccuracies of the model or sensors or simply due to the dynamic nature of the terrain. We provide a state space based framework for stabilisation of a high dimensional multi-legged robot which detects and recovers from unexpected events such as leg slip. We experimentally evaluate our approach using a modified PhantomX hexapod robot with extended tibia segments which significantly reduces its stability. Our results show that roll and pitch stability is improved by 2x when using the proposed method.