The coordinate system in which humans learn novel motor skills is controversial. The representation of sensorimotor skills has been extensively studied by examining generalization after learning perturbations specifically designed to be ambiguous as to their coordinate system. Recent studies have found that learning is not represented in any simple coordinate system and can potentially be accounted for by a mixed representation. Here, instead of probing generalization, which has led to conflicting results, we examine whether novel dynamics can be learned when explicitly and unambiguously presented in particular coordinate systems. Subjects performed center out reaches to targets in the presence of a force field, while varying the orientation of their hand (i.e. wrist angle) across trials. Different groups of subjects experienced force fields that were explicitly presented either in Cartesian coordinates (field independent of hand orientation), in Object coordinates (field rotated with hand orientation) or in Anti-object coordinates (field rotated counter to hand orientation). Subjects learned to represent the dynamics when presented in either Cartesian or Object coordinates, learning these as well as an ambiguous force field. However, learning was slower for the object-based dynamics and substantially impaired for the Anti-object presentation. Our results show that the motor system is able to tune its representation to at least two natural coordinate systems but is impaired when the representation of the task does not correspond to a behaviorally relevant coordinate system. Our results show that the motor system can sculpt its representation through experience to match those of natural tasks.
Significance Statement: The coordinate system in which humans learn motor skills has been highly controversial. Despite extensive experimental work, the results are highly conflicting (e.g. intrinsic joint-based vs. Cartesian vs. mixed coordinates etc). In our study we show that the motor system is able to sculpt and tune its representation to at least two natural coordinate systems. Importantly, we also show that learning is impaired for a task of equal complexity that does not correspond to a naturalistic coordinate system. Our results suggest that the previously conflicting findings primarily arise from the use of novel skills which are ambiguous as to their coordinates, thereby making the experimental results highly sensitive to small differences in the experimental design.
Authors report no conflict of interest.
We thank the Wellcome Trust (WT091547MA and WT097803MA) and the Royal Society (Noreen Murray Professorship in Neurobiology to D.W.) for support.