Reflexes occur always ``too late''; i.e., only after a (unpleasant, painful, dangerous) reflex-eliciting sensor event has occurred. This defines an objective problem which can be solved if another predictively acting sensor input exists. We present a new learning algorithm (ISO-learning) which performs a confounded-correlation between the primary reflex and a possible earlier reflex: the system learns the relation between a primary reflex and an earlier signal in order to create a new predictive reflex. To show this, we will embed ISO-learning into a behavioural system (a robot) thereby producing a closed loop situation where the learner's actions influence its own sensor inputs to the end of creating an autonomous agent. We will demonstrate that ISO-learning can successfully solve the classical obstacle avoidance task in correlating a reflex behaviour (retraction after bump) with signals of range finders (turn before the bump).