Abstract. This paper explores the capabilities of continuous time recurrent neural networks (CTRNNs) to display reinforcement learning-like abilities on a set of T-Maze and double T-Maze navigation tasks, where the robot has to locate and "remember" the position of a reward-zone. The "learning" comes about without modifications of synapse strengths, but simply from internal network dynamics, as proposed by [12]. Neural controllers are evolved in simulation and in the simple case evaluated on a real robot. The evolved controllers are analyzed and the results obtained are discussed.

Abstract. In [8] Yamauchi and Beer explored the abilities of continuous time recurrent neural networks (CTRNNs) to display reinforcementlearning like abilities. The investigated tasks were generation and learning of short bit sequences. This "learning" came about without modifications of synaptic strengths, but simply from internal dynamics of the evolved networks. In this paper this approach will be extended to two embodied agent tasks, where simulated robots have acquire and retain "knowledge" while moving around different mazes. The evolved controllers are analyzed and the results are discussed.

Two classes of dynamical recurrent neural networks, Continuous Time Recurrent Neural Networks (CTRNNs) (Yamauchi and Beer, 1994) and Plastic Neural Networks (PNNs) (Floreano and Urzelai, 2000) are compared on two behavioral tasks aimed at exploring their capabilities to display reinforcement-learning like behaviors and adaptation to unpredictable environmental changes. The networks report similar performances on both tasks, but PNNs display significantly better performance when sensory-motor re-adaptation is required after the evolutionary process. These results are discussed in the context of behavioral, biological, and computational definitions of learning.

No abstract