The neural network encodes trajectories by using competitive and temporal Hebbian learning rules and operates by producing the current and the next position for the robotic arm. Different types of trajectories can be learned independently of their complexity. Tests will focus on trajectories with one crossing point. The algorithm is able to reproduce the trajectories accurately and unambiguously due to context units used together with the input. Also, the proposed model is shown to be fault-tolerant and can respond well in the presence of noisy inputs.
A new feature of the navigation learning approach ist that the reinforcement signal from the environment is represented through reward and penalty surfaces to endow the agent with the ability to plan and to behave reactively. The agent solves the goal-directed reinforcement learning problem in which a first learning stage finds a path, based only on local information, and this path is a meliorated through further training. The proposed task is executed in an initially unknown environment, then, the initial viable solution is improved, employing a variable learning rate for the reward evaluation. The simulations suggest that the agent always reach the target, even in complex environments.