Table 1.

Summary of ILGA components and results

Component	Result
Object feature population codes	Aids generalization between objects with slightly different features (i.e., objects of a similar size or orientation)
Self-organizing map (SOM)	Aids generalization between objects with similar combinations of features and novel objects with feature combinations similar to those seen before
Modulation of SOM learning rate by reinforcement	Causes SOM to preferentially represent feature combinations of successfully grasped objects
Dynamic neural fields (DNFs)	Allow selection of motor parameters based on input strength
Noise in DNF activity	Promotes exploration of parameter space
Learned connection weights between AIP and premotor DNFs	Bias the selection of reach and grasp parameters to successfully grasp object
Learned connection weights between premotor DNFs	Bias the selection of reach and grasp parameters that depend on values of other parameters (i.e., the appropriate wrist rotation depends on the chosen reach offset)
Fixed connection weights between V6a/MIP and premotor	Allow accurate reaches. Future versions could change with learning to simulate reach learning
Dynamic movement primitives (DMPs)	Implement reach trajectory planning in a way that is easily extended to handle more complex trajectories