Abstract: A novel biologically inspired controller for the autonomous
navigation of a mobile robot in an evasion task is
proposed. The controller takes advantage of the environment by
calculating a measure of danger and subsequently choosing the
parameters of a reinforcement learning based decision process.
Two different reinforcement learning algorithms were used: Qlearning
and Sarsa (λ). Simulations show that selecting dynamic
parameters reduce the time while executing the decision making
process, so the robot can obtain a policy to succeed in an escaping
task in a realistic time.