Figure 2.

(a) Subjects undergoing testing using the upper extremity robot first develop internal action models by training to the task in the left workspace (1). They then demonstrate generalization of the internal action models by performing the task in the right workspace. The performance of the task may be done either in visual coordinates that were similar to the training (2) or in proprioceptive coordinates that were similar to the training (3), allowing differential examination of the contribution of visual and proprioceptive feedback during the development of the internal action model. (b) Children with autism and typically developing (TD) children have similar performance during the training (target 1). Children with autism show significantly more proprioceptively guided generalization (to target 3) in contrast to visually guided generalization (target 2). (c) Generalization to proprioceptive coordinates (target 3) correlates with Autism Diagnostic Observation Schedule–General (ADOS-G) score in autistic subjects (r = 0.572, P = .032). ASD = autism spectrum disorder. (d) In all children (blue dots = TD children; red dots = autistic children), relative generalization to proprioceptive coordinates (target 3) versus visual coordinates (target 2) correlates negatively with imitative function (r = -0.57, P = .006). This figure first appeared in a modified form in Nature Neuroscience (Haswell and others 2009); reprinted with permission.