A framework for the acquisition of perception-based navigational behaviors by autonomous mobile robots is presented. A globally coupled chaotic system is applied for learning of reactive action-rules. The control architecture of a robot consists of chaotic elements each encoding an action-rule. The elements have the dynamics, which are designed so that the elements can collectively execute reinforcement learning. The proposed method has been tested in simulations dealing with a navigation task in changing environments. Simulation results indicate that the robot acquired behaviors such as light-seeking and wall-following. The robot became able to move along a wall with its head swinging reciprocally so as to keep perceiving the wall on its left side. This strategy is flexible enough to enable the robot to search for, keep following, and turn along the wall. The process of behavior acquisition is also discussed from the point of view of credit assignment.