Learning multiple distinct patterns in a task of rhythmic interlimb coordination

A basic model for learning new patterns in a task of performing rhythmic movement of two limbs with same frequency, but shifted in phase by a set amount, the relative phase φ, was described in the report “A dynamic field model of learning new patterns in rhythmic interlimb coordination”, from May 29th, 2007. The model consists of three separate, interconnected parts, command, coordination and feedback, forming a closed loop. The three layered command field develops self-sustained peaks, which change the coordination dynamic, creating an attractor at the activated site. The feedback field compares the produced to the desired results, and merges that information with the memory layer from the command field to create corrective input. Learning takes place in the form of preactivation of field sites, or preshape. Whenever a field site supports an active peak, preshape is increased at that sites, thus forming preshape peaks at those sites that tend to produce the best results, which in turn support the quick formation of peaks at those sites in subsequent trials, and help stabilizing these peaks by providing additional input. The model gave a good description of the process of learning to produce a single new relative phase. The predictions of the model were qualitatively consistent with the experiment results in terms of mean produced relative phase, and stability. In the state it is described here, the model has been enhanced in two major ways. First, the dynamic feedback loop was closed. In the earlier version, feedback was only given in large intervals, between trials. Now feedback is given on a similar, but slower timeframe as the command and coordination dynamics run. While seemingly a cosmetic change, the new model shows far nicer behaviour in several aspects, as changes that were previously limited to the switch between trials can now occur at different times and with varying frequency.