Flexible strategies for sensory integration during motor planning

Abstract

When planning target-directed reaching movements, human subjects combine visual and proprioceptive feedback to form two estimates of the arm's position: one to plan the reach direction, and another to convert that direction into a motor command. These position estimates are based on the same sensory signals but rely on different combinations of visual and proprioceptive input, suggesting that the brain weights sensory inputs differently depending on the computation being performed. Here we show that the relative weighting of vision and proprioception depends both on the sensory modality of the target and on the information content of the visual feedback, and that these factors affect the two stages of planning independently. The observed diversity of weightings demonstrates the flexibility of sensory integration and suggests a unifying principle by which the brain chooses sensory inputs so as to minimize errors arising from the transformation of sensory signals between coordinate frames.

Figures

Figure 1. Two types of reaching errors induced by shifted visual feedback

(a) Error in planned movement direction resulting from a leftward error in the position estimate used to plan the movement vector (“MV error”). This error is caused by a leftward shift of the visual feedback. The planned movement direction for a reach to the target at 90° (blue dot) differs from the true hand-to-target direction, resulting in a clockwise error (blue arrow). (b, c) Error in initial movement direction resulting from a leftward error in the position estimate used to evaluate the inverse model of the arm and compute the motor command (“INV error”). The wrong joint angle displacement (motor command) is planned for the desired movement vector (b), resulting in a movement error (c). (d) The bimodal MV error pattern resulting from a leftward error in estimated position. The blue arrow represents the MV error shown in (a). (e) A leftward error in estimated position results clockwise errors for all target directions. (f) Measured errors in initial direction are a combined effect of MV and INV errors (black curve). A rightward error in position estimates would produce an error pattern of opposite sign (gray curve). Here and in subsequent figures, 0° refers to a rightward vector and positive angles are counterclockwise.

Figure 2. Predicted changes in Experiments 1 and 2

Each panel shows the movement direction errors predicted for different values of αMV and αINV , which represent the relative weighting of visual feedback at the first and second stages of reach planning, respectively (see Methods) . Both panels include the model predictions for αMV = 0.9, αINV = 0.3 (black lines), corresponding to the values expected for reaches to a visual target with fingertip feedback. (a) Proprioceptive targets in Experiment 1 are expected to result in a lower value of αMV. (b) Arm feedback in Experiment 2 is expected to result in a higher value of αINV.

Figure 3. Data collection and task parameters

(a) Virtual visual feedback setup. (b) In Experiment 1 a visual and a proprioceptive target were present on every trial; the “go” signal cued subjects which reach to make. Visual feedback of the fingertip position was either veridical (gray dot) or shifted leftward (white dot) or rightward by 6 cm. (c) In Experiment 2, subjects reached to visual targets with visual feedback that showed either the fingertip or a simple, polygonal rendering of the arm. Visual feedback either reflected the true position of the fingertip or was shifted leftward (white dot, polygon) or rightward by 7 cm.

Figure 4. Sample data, Experiment 1

(a) Raw reach trajectories for a representative subject. Filled symbols, reach targets; corresponding open symbols, reach endpoints. (b) Initial reach direction with respect to target direction, using the color conventions of (a). (c) Reach errors induced by visual feedback shifts, computed by subtracting the baseline directional biases (dotted lines in b) from the initial directions in the shifted conditions (solid lines in b). Model fits, dashed lines. Best-fit weighting parameters: αMV = 0.96, αINV = 0.32 for visual-target reaches and αMV = 0.49, αINV = 0.19 for proprioceptive-target reaches. Error bars, ±1 s.e.m.

Figure 5. Group data, Experiment 1

(a) Average initial direction (±1 s.e.m.), with respect to target direction, for visual- and proprioceptive-target reaches. (b) Average errors in initial direction (±1 s.e.m.). Color conventions as in Figure 4.

Figure 6. Model fits, Experiment 1

(a,b) Fit values of αMV and αINV for reaches to visual and proprioceptive targets, respectively. Ellipses represent 1 s.e.m.. Dotted lines, αINV = αMV and αINV = 0. (c) Differences in model fit between visual- and proprioceptive target reaches. The top left half of the symbol is filled if the task-dependent change in αMV is significant (permutation test, P < 0.05), and the bottom right half is filled if the change in αINV is significant. +, average across subjects. In all plots, the gray symbols are for the example subject shown in Figure 4.

Figure 7. Sample and group data, Experiment 2

(a,b) Data from a representative subject in the fingertip-feedback and arm-feedback conditions, respectively. Gray, rightward visual shift; black, leftward visual shift; dashed lines, mean error across the six target directions. Error bars, ±1 s.e.m.. Arrows indicate ΔE̅, the difference between the mean errors for rightward- and leftward-shifted visual feedback. (c) Differences in mean errors for all subjects. Gray dot, example subject whose data are shown in (a) and (b); dashed line, ΔE̅Arm = ΔE̅Fingertip.

Figure 8. Model fits, Experiment 2

(a,b) Fit values of αMV and αINV for fingertip-feedback and arm-feedback trials, respectively. (c) Differences in model fit between arm- and fingertip-feedback reaches. In all plots, the gray dot represents the subject whose data is shown in Figure 7a and b. Other plotting conventions as in Figure 6.