Estimating the sources of motor errors for adaptation and generalization

Abstract

Motor adaptation is usually defined as the process by which our nervous system produces accurate movements while the properties of our bodies and our environment continuously change. Many experimental and theoretical studies have characterized this process by assuming that the nervous system uses internal models to compensate for motor errors. Here we extend these approaches and construct a probabilistic model that not only compensates for motor errors but estimates the sources of these errors. These estimates dictate how the nervous system should generalize. For example, estimated changes of limb properties will affect movements across the workspace but not movements with the other limb. We provide evidence that many movement-generalization phenomena emerge from a strategy by which the nervous system estimates the sources of our motor errors.

Figure 1

Simulated intralimb generalization. (a) A velocity-dependent force field in the right region of the workspace is adapted to. (b) Reaches are made in the unexperienced left region of the workspace to test for generalization. (c) A disturbance is estimated as an entirely extrinsic velocity-dependent force field and the arm is exposed to an intrinsic field (top) and an extrinsic field (bottom), (d) A disturbance is adapted to as misestimates in limb properties and the arm is exposed to an intrinsic velocity-dependent force field (top) and an extrinsic field (bottom), (e) Experimental results for intralimb generalization (reproduced from Shadmehr & Mussa-Ivaldi, ‘94) when subjects are exposed to an intrinsic and extrinsic force field, top and bottom respectively. (f) Simulated intralimb generalizations showing a mixture of inferred body and world estimates.

Figure 2

Simulated Interlimb Generalization. (a) Adaptation and (b) interlimb generalization with a curl field. (c) Experimental evidence (reproduced from Criscimagna-Hemminger et. al. ‘03) showing early exposure, adaptation (late training) and interlimb generalization (test) in a clockwise (Extrinsic) and counter-clockwise (Intrinsic) field. (d) Simulated data with our model demonstrating similar findings as the result of a small estimated external field and large misestimates in limb parameters. (e) Increasing the body’s uncertainty (decreasing α to 0.1) our model predicts asymmetric generalization. The top plots displays generalization from the dominant to the nondominant limb, while the bottom plots displays generalization from the nondominant to the dominant limb with little transfer of the adapted skill.

Figure 3

Simulated Coriolis room generalization. (a) A Coriolis force disturbance is adapted to and (b) interlimb generalization is observed. (c) Experimental results of motor adaptation and interlimb generalization reproduced from DiZio & Lackner ‘95. (d) Simulated experiment predicting similar results.

Figure 4

Simulated inertial perturbation generalization. (a) An inertial disturbance is adapted to with the dominant limb and (b) interlimb generalization is observed with the nondominant limb. (c) Experimental results for initial exposure reaches with nondominant left arm with out the mass, with a mass before, and after training with the dominant right arm (reproduced from Wang & Sainburg ‘04.). (d) Simulated experiment demonstrating similar predictions for initial reaches with nondominant arm.

Figure 5

Comparison of after effects. Maximum perpendicular errors from a straight line are plotted versus trials. (a) Experimental data displaying reach errors (mean of 6 subjects) during adaptation (left portion) and after effects with and without holding a robot handle, black and grey lines, respectively (reproduced from Cothros et. al. ‘06) (b) Simulated data of the same experiment (mean of 8 simulated runs). When the simulated limb let’s go of the handle, world estimates are neglected. (c) Simulated data with the two-module source estimation model.