The Long-Term Absence of Static Stereopsis Cultivates Adaptive Planning of Reaching-to-Grasp

Abstract

Purpose

To determine how the absence of stereopsis affects sensorimotor adaptation in visually guided reaching-to-grasp by comparing slant perception and depth-directed motor control between adults with long-term stereoblindness and typically sighted adults.

Methods

Twenty-four adults (12 stereoblind, 12 typically sighted) completed a slant-matching task (perceptual estimate of surface orientation) and a reaching-to-grasp task (motor execution in depth). Outcomes indexed perceptual accuracy and grasp kinematics, including approach speed and the timing of grip alignment to surface slant.

Results

Groups performed comparably on slant matching, but the stereoblind group adopted distinct grasping strategies, characterized by a faster approach to the target and delayed grip alignment relative to controls.

Conclusions

Findings indicate adaptive reorganization of motor planning following long-term deprivation of static binocular (disparity) cues. The visuomotor system appears to recalibrate the balance between feedforward and feedback control-placing greater weight on real-time sensory information when anticipatory planning is constrained-highlighting flexible, ecological pathways for development and adaptation.

Stereo vision refers to the ability to compute depth based on the binocular disparity between the retinal images of an object in the left and right eyes.¹ Compared to herbivores, the evolution of frontally positioned eyes holds particular significance for humans as research has demonstrated that stereo vision plays a crucial role in reaching-to-grasp movements, as monocular observation is typically associated with reduced accuracy and lower peak velocities.² However, the precise mechanisms through which stereoscopic vision facilitates reaching-to-grasp movements have long been a subject of debate. Some hypotheses posit that binocular information enhances depth perception, thereby supporting both the planning and execution of natural reaching-to-grasp actions.^2,3 However, empirical evidence regarding this hypothesis remains inconclusive.^4–8,53 For instance, Watt and Bradshaw⁵ argued that the absence of binocular vision primarily degrades critical spatial information necessary for visuomotor processing, rather than merely introducing biases in the estimation of an object's distance or size. Greenwald et al.⁸ also reported that binocular cues contributed more to online control than planning of reaching.

To investigate how stereo vision benefits reaching movement, Bingham et al.⁷ proposed the online disparity matching theory to explain how accurate reaching movements are achieved by continuously tracking the binocular disparity of the hand while maintaining focus on the target, without perceiving the absolute distance of the target beforehand (as illustrated in Fig. 1a). Building upon this framework, Anderson et al.¹⁰ and Bingham et al.¹¹ later integrated principles from tau models⁹ to develop a refined approach incorporating a novel informational variable, disparity tau (τ_α), as well as a proportional rate control strategy. This approach allows for maintaining a consistent proportion between τ_α and its rate of change $\left({\dot{\tau }}_{\alpha }\right)$ to guide reaching movements. Further work has confirmed this model.^12,13

Figure 1.

( a) The change in hand disparity from t₁ (left) to t₂ (right) as the hand moves toward the target. According to the proportional rate control strategy, the binocular disparity of the hand gradually decreases and ultimately reaches zero upon contact with the target. This mechanism suggests that accurate reaching movements do not require pre-estimating the target's distance. Instead, the movement is guided by continuously tracking the disparity of the hand while maintaining focus on the target. (b) The concepts of CDOT (left panel) and IOVD (right panel). In CDOT, the observer first processes spatial information by detecting the binocular disparity of images on the retinas and then tracks how this disparity changes over time, incorporating temporal information. In contrast, IOVD follows the opposite sequence, with the observer initially processing temporal information by computing the derivative of optic flow and then comparing spatial information by assessing the differences between the two optic flow derivatives across the eyes. (c) Schematic illustration of the IOVD mechanism. When the hand is closer to the observer, the velocity vector of the moving hand projects differently onto each retina. The vector can be decomposed into components orthogonal and parallel to each eye's line of sight. Assuming the hand is reduced to one point, because motion components parallel to the line of sight cannot be detected by the visual system (as the moving dot does not produce a change in image size), only the orthogonal (tangential) components remain available to provide differential velocity signals between the two eyes. The blue and red dashed lines denote the lines of sight from the left and right eyes, respectively, whereas the corresponding arrows indicate the tangential components of the velocity vector projected to each eye. From the left to the right, the IOVD changes with the hand motion.

On the other hand, stereoblindness, a lasting difficulty in perceiving depth through stereoscopic cues, affects 4% to 14% of the population.^14,15 Moreover, more than 30% of the population exhibit some degree of binocular dysfunction.¹⁶ Consequently, it raises a question how individuals with such a dysfunction successfully adapt their visually guided reaching to maintain motor function. This study investigates how long-term sensory deprivation influences the development of visuomotor strategies, offering new insights into the mechanisms by which visual input influences the sensorimotor control.

It is important to note that stereoblindness does not necessarily imply a complete absence of depth perception from stereo cues. Individuals classified as stereoblind, typically based on static random-dot stereogram tests, may still access motion-based depth information. Depth perception from motion primarily relies on three channels. The most basic is optic flow, the pattern of motion in the visual scene as the observer moves, which can be perceived monocularly and provides depth (particularly change in depth) through velocity gradients.¹⁷ When both eyes are available, two additional binocular channels contribute: Change of Disparity Over Time (CDOT) and Interocular Velocity Difference (IOVD).^18–21 CDOT involves tracking changes in binocular disparity over time, whereas IOVD compares differences in monocular optic flow velocities between the two eyes, offering depth cues based on motion velocity differences (see Fig. 1).

Theoretically, because stereoblind individuals have problem perceiving static stereograms and CDOT relies on image-based disparity, they likely lack access to CDOT while potentially using IOVD for stereo motion perception to maintain stability. This idea is supported by Nefs et al.,²¹ who found significant individual differences in motion-in-depth (MID) sensitivity: although some participants relied on both CDOT and IOVD, others relied exclusively on one mechanism. Importantly, individuals relying solely on IOVD exhibited poor or absent static stereoacuity, resembling stereoblind individuals, suggesting that alternative motion-based cues may compensate for impaired disparity processing. Further supporting this, Tidbury et al.²² showed that even those without measurable stereoacuity could experience vivid depth “pop-out” effects when viewing three-dimensional (3D) stereoscopic videos.

Building on these foundations, the present study investigates how the visuomotor system develops and adapts in response to the prolonged absence of static stereoscopic input. We recruited 24 participants, 12 stereoblind individuals (stereoblind group) and 12 individuals with typical stereovision (the control group), to perform both a slant matching task and a reaching-to-grasp task. We hypothesize that the absence of static stereopsis will lead stereoblind participants to rely on the IOVD channel during the reaching-to-grasp task. This reliance may be reflected in higher movement speeds, as previous research has shown that the IOVD channel is particularly effective at processing depth information at higher motion speeds.^23–25,39 Regarding slant perception, we expect the findings to be consistent with our previous study, exhibiting no significant differences.²⁶

Material and Methods

Participants

A priori power analysis (using G*Power) indicated that a total sample size of 24 participants would be sufficient to detect a medium-sized interaction effect (f = 0.25) in a repeated-measures ANOVA with five conditions and two groups, assuming α = 0.05 and power = 0.85.

Twenty-four subjects (20 females, four males; mean age = 22.5 ± 2.2 years) took part in the experiment, including 12 stereoblind individuals and 12 stereo-normal controls. All participants were right-handed and had normal or corrected-to-normal visual acuity (20/20), except that two stereoblind participants with childhood anisometropia did not use corrective lenses; their good eye exhibited near-normal acuity (around 20/30), whereas the other eye showed reduced acuity (around 20/80). All participants were naïve to the purpose of the study. Visual acuity was assessed with participants wearing their habitual correction; uncorrected (naked-eye) vision was not tested. Stereoblindness was first assessed with the Randot Stereotest (Stereo Optical Co., Inc., Chicago, IL, USA) at a threshold of 400 arcsec and then confirmed with a computer-based stereoacuity task. All stereo-normal participants demonstrated thresholds as fine as 40 arcsec. Among participants in the stereoblind group, most were unaware of their condition and reported no difficulties in everyday object interaction, and four individuals noted challenges in perceiving depth in 3D movies; the majority of cases were attributed to childhood anisometropia, with one participant reporting a history of surgically corrected strabismus.

Apparatus and Stimuli

Half of the surfaces used for judgment were smooth and half were textured using Voronoi patterns adapted from Saunders and Chen.²⁷ The setup included an adjustable slant platform and a laptop for trial control and data recording (Fig. 2a). The experiment was programmed in MATLAB, with each object coded so that during the experiment, the screen displayed the object's number and slant degree for each trial. This allowed the experimenter to select and position the correct object at the assigned slant. Kinematic data were recorded using the trakSTAR 3D Guidance electromagnetic tracking system (Ascension Technology Corporation, St. Louis, MO, USA). The system sampled at 240 Hz, providing a positional accuracy of approximately 1.4 mm root mean square (RMS) and a rotational accuracy of 0.5° RMS. Two miniBIRD sensors, attached to each side of the platform, recorded the actual slant degree.

Figure 2.

( a) Participants reported their slant perception using their index finger and thumb to hold a small comparison disk shaped surface. The coordinate system, set up for a seated participant facing the table, was defined as follows: the x-axis runs horizontally across the table from the participant's left to right, the y-axis points vertically upward, perpendicular to the table surface, and the z-axis extends horizontally toward the participant's front. In this setup, the frontal plane is at a slant of zero degrees, while the transverse plane is at a slant of 90°. All coordinates mentioned in this article follow this reference frame. (b) The procedure in the reaching-to-grasp task. Before each trial, the experimenter placed a target object surface and adjusted it to a specific slant. Subsequently, the experimenter pressed the space key, producing a beep to signal the participants to open their eyes and perform the reaching-to-grasp task. (c) A representative reach trajectory. (d) Illustration of grip inclination. (e) Illustration of grip aperture. (f) Four shapes of targets used in this study.

Experiment Design and Procedures

This study included two tasks. The first was a slant matching experiment using the method of adjustment to assess slant perception with actual object surfaces (Fig. 2a). The second was a reaching-to-grasp task, in which we recorded and analyzed participants’ reach trajectories as they grasped objects positioned at a slant (Fig. 2b). Both experiments manipulated four variables: target shape (foiur types, as shown in Fig. 2f.), texture presence (with/without) slant (five levels: 35°–55°, increment = 5°) angle, and vision (monocular/binocular) mode. The selection of the four shapes was informed by previous work.^28–30 In monocular condition, the observers worn an eye-patch and performed the task using their dominant eye. Each condition was repeated four times, totaling 320 trials across two randomized, counterbalanced blocks (one monocular, one binocular).

Experiment 1: Slant Matching

The slant matching task was similar to that of Cherry and Bingham,³¹ matching optical texture and contour, shapes and texture on one of these, where participants viewed actual object surfaces at a slant and matched the slant by aligning a comparison object by hand. In this experiment, however, half of the participants were stereoblind. To eliminate the ability to match projected optical patterns, the reference disk was textureless and round in shape, distinguishing it from the target. Target surfaces were created by laser-cutting plexiglass, with a LEGO piece attached to the back for easy mounting on a supporting surface. During each trial, the experimenter placed a slanted flat surface on an adjustable platform, and participants matched the slant using a small hand-held disk. The dependent variable was the slant of the comparison disk (Fig. 2a).

After signing consent and receiving instructions, participants sat at a table with the judgment platform. At the start of each trial, participants closed their eyes while the experimenter adjusted the target's slant, then pressed a key, prompting a beep to signal participants to open their eyes and perform the task. Each participant aligned a comparison disk in their right hand to match the slanted target. Participants were able to see both the target and the reference disk. The disk was located ≈20 cm to the right of the target ≈20 cm closer to the participant, and at the same height. If the disk was 5 cm away from this location, a warning prompted adjustment. Two miniBIRD sensors recorded the disk's slant. After each judgment, participants pressed a button with their left hand to proceed to the next trial and closed their eyes.

Experiment 2: Reaching to Grasp

This experiment examined hand movement trajectories during reaching-to-grasp actions with objects at various slants. Participants grasped and lifted the target object with their index finger and thumb, then placed it in a specified location. The same participants, stimuli, and apparatus from Experiment 1 were used, with the only difference being the reaching-to-grasp task instead of slant matching.

As in Experiment 1, participants closed their eyes before each trial while the experimenter set up the object and adjusted its slant. After a beep signal, participants raised their right hand, opened their eyes at a second beep, raised their hand directly upwards from a start position on the table, then reached to grasp the target with their index finger and thumb, lifting it and returning it to the start position which was an 8 × 8 cm square located 25 cm from the center line at the table's edge (approx. coordinates to the target: 25, −20, −20 cm). After placing the object, they closed their eyes to begin the next trial. Note: the contact locations of the index and thumb on the target were not constrained or specified to the participants who were therefore free to grasp it as they chose. In Experiment 2, the finger and thumb started in a pinched position and rest in a LEGO block. The design, conditions, and number of trials matched those in Experiment 1. Participants were instructed to perform the task in a manner consistent with their everyday behavior, without receiving any additional task-specific instructions.

Statistical Analysis

All data were inspected visually trial by trial to ensure quality. We mainly used SPSS and MATLAB to do data analyses. Data and scripts are openly accessible (see the link in Statements).

For the slant matching task, we conducted a mixed-design ANOVA using SPSS to test the degree of slant (five levels, 35°–55°), viewing conditions (binocular/monocular) and group effect on slant perception in physical environment. Statistical design for this experiment can be found in the Results describing Figure 3a.

Figure 3.

( a) This figure presents the results of the slant matching task for two groups (stereoblind and control) across four cue type conditions. Linear regression lines are fitted, with shading indicating standard errors. Two eyes with texture: Both groups show strong linear relationships between estimated and actual slant (R² ≈ 0.86–0.88). Two eyes without texture: Slightly lower R² values (0.79–0.89). One eye with texture: Weaker correlations (R² = 0.38–0.47). One eye without texture: Moderate correlations (R² = 0.48–0.56). (b) The normalized time course of velocity during the reaching phase. (c) The peak Velocity of stereoblind group is significantly greater when the stereo information is available. Error bars represent standard errors of the mean.

For the reach-to-grasp experiment, raw data was first filtered by MATLAB in-built function smoothdata. To use smoothdata, we determined the size of a moving window (the window size is 10% of the data length), which slides down the length of the entry vector and computes an average over the elements within each window. To investigate the group effect, we analyzed the initiation times (IT), movement times (MT), peak velocity (PV, Fig. 3c), max grip aperture (MGA, Fig. 4b) for both groups using mixed-design ANOVA.

Figure 4.

Grip aperture. (a) The results of dynamic grip aperture as a function of time. The group was not significant under the four conditions. (b) Max grip aperture of the two groups under four cue type conditions. Error bars represent standard errors of the mean. The vision was significant.

To examine the group effect on dynamic change of velocity (Fig. 3b) and GA (Fig. 4a) as functions of normalized time, the time course for each trial was divided into 40 equal segments, with average measurement computed per segment. We used MATLAB for these analyses. To evaluate the similarity in the shape of velocity profiles between groups, we used Pearson correlation analysis.

To examine the effect of slant of target on grip inclination, we analyzed both the distance-dependent and time-dependent evolution of grip inclination. Hierarchical linear models (HLM, see more detail in Results) were used to analyze these data.

Results

Matching Task

This experiment aimed to explore the role of stereo vision in perceiving the slant of real object surfaces. The results were displayed in Figure 3a. A mixed-design ANOVA across slants revealed a significant main effect of cue type (F(3, 66) = 2.856, P = 0.04, η² = 0.115 and a significant cue type × group (F(3, 66) = 3.670, P = 0.016, η² = 0.143), but no main effect of group (F(1, 22) = 0.006, P = 0.9, η² < 0.001). Post hoc one-way ANOVAs showed a significant group difference in the two-eyes-with-texture condition (F(1,22) = 9.893, P = 0.005, η² = 0.310) with no significant differences in the other conditions (F(1, 22) < 3.4, P > 0.07, η² < 0.270).

Overall, both groups performed similarly in the slant matching task, consistent with Yang et al.²⁶ However, a significant group effect in the two-eyes-with-texture condition suggests that control participants benefited more from binocular observation when texture was present. This implies that texture enhances stereoscopic perception by amplifying the advantage of binocular cues. The synergistic interaction between texture and stereo cues likely provides richer slant information than either alone, with texture contributing not only a monocular gradient but also an additional disparity gradient. Although the combined influence of texture and stereo cues on depth perception has been extensively studied, it remains a topic of debate.^28,32,33

Grasping Task

Raw 3D position data were recorded from sensors on the thumb and index finger. Trajectories were smoothed using MATLAB's smoothdata function. Trial-by-trial visual inspection was conducted to ensure data quality before detail analyses. Of the 7680 trials, 59 were excluded because of incomplete movements caused by incidental disruptions, resulting in 7621 trials retained for analysis.

Figure 2c shows a representative reach trajectory. After stimulus onset, a beep signaled participants to raise their right hand, followed by another beep to open their eyes and begin movement. We first quantify movement timing by estimating initiation time (IT) and movement time (MT) for each participant and comparing these measures across groups. We then characterize kinematics by analyzing velocity profiles over normalized time and comparing peak velocity (PV) between groups. To further describe the grasp component, we present the time course of grip aperture (GA) and compare maximum grip aperture (MGA) across groups. Finally, we examine the evolution of grip inclination (GI) as a function of absolute hand–target distance.

Initiation Time and Movement Time

IT, defined as the time from initial hand elevation to 5% of peak velocity, averaged 0.709 ± 0.22 second for the control group and 0.718 ± 0.23 second for the stereoblind group. A mixed-design ANOVA revealed no significant main effect of group, vision, or their interaction (F(1, 22) < 0.1, P > 0.8, η² < 0.01). MT, measured from the end of initiation until deceleration to 5% of peak speed, averaged 0.80 ± 0.12 second for the control group and 0.82 ± 0.16 second for the stereoblind group, showed no significant group (F(1, 22) = 0.037, P = 0.848, η² = 0.002), but vision (F(1, 22) = 4.805, P = 0.039, η² = 0.179), indicating faster movements under binocular viewing. Texture had no significant effect nor the interactions (F(1, 22) < 0.007, P > 0.1, η² < 0.02). In summary, long-term stereoblindness did not alter movement timing during reaching-to-grasp actions.

Time Course of Velocity

We analyzed whether the groups differed in hand movement velocity, particularly in PV. Velocity (via two-point differentiation) was measured at the center point between the index finger and thumb, and the time course for each trial was divided into 40 equal segments, with average velocity computed per segment (Fig. 3b) to allow computation across trials. To evaluate the similarity in the shape of velocity profiles between groups, we used Pearson correlation analysis. The results revealed a strong positive correlation between the temporal velocity profiles of the two groups under four cue type conditions (r ≥ 0.830, P < 0.001), indicating that the two groups exhibited highly similar temporal patterns in velocity trajectories.

We observed that the stereoblind group generally exhibited a higher PV than the control group, which aligned with our hypothesis. We then conducted a mixed-design ANOVA to compare PV across the four viewing conditions between groups. Note that PV here was calculated before time normalization. With time normalization, the velocity at a given time point represents the average velocity across a specified time window Figure 3b, so PV values here are higher than those shown in Figure 3b. As showed in Figure 3c, group was not significant (F(1, 22) = 2.078, P = 0.164, η² = 0.086). Vision was significant (F(1, 22) = 9.757, P = 0.005, η² = 0.307), suggesting a faster movement using two eyes. Texture was not significant (F(1, 22) = 0.803, P = 0.380, η² = 0.035). The interaction between vision × group was significant (F(1, 22) = 4.851, P = 0.038, η² = 0.181). Post hoc tests using one-way ANOVA showed that the group effect was significant with the binocular observation (F(1, 22) = 9.145, P = 0.004, η² = 0.166) and was not significant with monocular vision (F(1, 22) = 0.683, P = 0.413, η² = 0.015) implying that when two eyes were available, PV of the stereoblind group was higher than that of the control.

Time Course of Grip Aperture and Max Grip Aperture

In this study, GA is defined as the distance between the index finger and thumb (Fig. 2e), reflecting the size perception of the target. Pearson correlation analysis showed a strong positive association between the GA profiles of the two groups across the four cue type conditions (r ≥ 0.850, P < 0.001), indicating highly similar temporal patterns in GA trajectories. We then conducted a mixed-design ANOVA to test whether group has an effect on MGA. Results showed that the group was not significant (F(1, 22) = 2.772, P = 0.110, η² = 0.112). Vision was significant (F(1, 22) = 6.615, P = 0.017, η² = 0.231), whereas texture was not (F(1, 22) = 2.078, P = 0.164, η² = 0.086). No significant interaction was found (F(1, 22) ≤ 1.570, P > 0.223, η² < 0.067). The results were shown in Figure 4a. In sum, the two groups did not differ in grasp aperture scaling and this pattern is consistent with some previous findings. For instance, Grant and Conway³⁴ also reported no significant difference in MGA between control participants and individuals with amblyopia. One possible explanation is that, through a limited number of interactions with graspable objects during the practice session, the motor systems of both groups developed stable proprioceptive calibrations between object properties and hand aperture, resulting in similar scaling of MGA through learned sensorimotor mappings rather than online stereoscopic estimation.

Distance-Dependent Evolution of Grip Inclination

GI (illustrated in Figure 2d, the results are shown in Figure 5, also see Appendix A for detailed calculation) refers to the angle at which the fingers approach an object, reflecting slant perception. We defined GI as 0° when the index finger is positioned directly above the thumb in vertical alignment and 90° when the index finger is positioned directly in front of the thumb horizontally. Motivated by the absence of reliable stereo information in stereoblind participants, we hypothesized that their representation of slant would be weaker early in the reach, leading to later, distance-dependent adjustments of hand orientation. Therefore, we analyzed GI as a function of distance-to-target and identified the GI Differentiation Point (GID), defined as the distance at which GI begin to diverge by target slant. Prior to the GID, participants showed no GI differences, indicating a pre-planning stage. After the GID, participants adjusted GI according to target slant, suggesting that the GID during reaching offers insights into each group's action strategies.

Figure 5.

The grip inclination of the two groups as a function of absolute distance when performing with two eyes or monocular vision. Shaded areas indicate standard error. The red vertical lines represent the slant effect was significant at those distance points. With binocular vision, the control group's grip inclination differentiation point occurred at 0% of the normalized time whereas the stereoblind group's occurred at 18%. With monocular vision, the grip inclination differentiation points for the control and stereoblind group were 0% and 16%, respectively, suggesting that stereoblind individuals prepare to grasp objects accordingly later than the control group, relying more heavily on feedback obtained during the later phase of movement to adjust their hand positioning accurately. The similarity between GI under binocular and monocular conditions within each group may reflect a transfer of calibration—participants may have adjusted their responses in the monocular condition based on prior experience under binocular viewing. The green shadowed area represents the when the interaction of group × slant was significant in each graph.

For each group, GI from the textured and textureless conditions was collapsed within viewing mode (binocular/monocular), as no significant effect of texture was observed. Figure 5 depicts GI for both groups across five target slants as a function of absolute distance to the target, under binocular and monocular viewing. The mean starting distance was 45.78 ± 4.08 cm away from the target. To standardize the analyzed approach segment across trials, analyses were restricted to the 40-cm interval preceding contact (i.e., absolute hand–target distance from 40 to 0 cm). Trials with initial distances less 40 were excluded (583/7,621; 7.6% trials were excluded).

The relationship between target slant and the progression of GI with distance was tested using HLMs fit separately within 50 distance intervals across the 40-cm trajectory. For each distance interval, GI was predicted from target slant and group (stereo-normal vs. stereoblind), with a random intercept for subject:

$$\begin{array}{ccc}\hfill G{I}_{i,t}& \sim & {\beta }_0+{\beta }_1\left(slan{t}_{i,t}\right)+{\beta }_2\left(Grou{p}_t\right)\hfill \\ & +& {\beta }_3\left(Grou{p}_t\times slan{t}_{i,t}\right)+{\in }_{i,t}\hfill \end{array}$$

where GI_i,t is the grip inclination for participant i at distance interval t, β₁ is the effect of slant, β₂ is the effect of group, β₃ is the effect of the interaction of slant × group, ∈ _i,t is the residual error.

The HLM (see Appendix B for more details) analyses revealed that, under binocular viewing, the Group first reached significance at 78% of the trajectory (β = 2.52, P = 0.01) and remained significant thereafter. The Slant emerged at 4% (β = 0.218, P = 0.01). A significant Group × Slant interaction appeared at 88% (β = 0.087, P = 0.04), indicating that the influence of slant on GI diverged between groups late in the reach. Under monocular viewing, the Group became significant at 94% of the trajectory (β = 2.41, P = 0.04) and persisted through the terminal phase. The Slant reached significance at 12% (β = 0.174, P = 0.003), whereas the Group × Slant interaction was not significant at any distance (β < 0.1, P > 0.07). Collectively, these results indicate that binocular viewing affords earlier slant-based guidance and reveals late-phase, slant-dependent group differences that are not detectable under monocular viewing.

We then quantified within-group slant effects on GI as a function of distance-to-target. Under binocular viewing, the control group exhibited GID at the first distance window (0% of the 40 cm trajectory; β = 0.147, P = 0.02), whereas the stereoblind group showed a delayed onset at 18% of the trajectory (β = 0.231, P = 0.024). Accordingly, the GID was 0% for controls and 18% for stereoblind participants. Under monocular viewing, GID was likewise 0% for the control group (β = 0.110, P = 0.02) and 16% for the stereoblind group (β = 0.152, P = 0.015).

We also plot the GI as function of normalized time for the two groups under both binocular and monocular condition to examine the effect of slant on GI, as shown in Figure 6. Additionally, we observed that the stereoblind participants initiated movements with approximately 5° greater slant in the monocular conditions compared to in the binocular conditions. A trial-by-trial analysis of the GI data indicated that this discrepancy was primarily driven by greater individual variability among stereoblind participants (see Appendix C). Notably, before initiating the grasp, participants closed their eyes, eliminating any visual input that could influence their initial hand position; therefore the starting slant should not reflect perceived slant of the target but rather individual preferences for initial hand positioning. As such, it is unlikely to indicate systematic perceptual or motor biases and does not have substantive implications for interpreting our findings.

Figure 6.

The GI as a function of normalized time for both groups under binocular and monocular conditions. It is apparent that the control subjects showed GID before the acceleration and the stereoblind subjects showed the GID near the end of acceleration. Within each group, the similarity in GID between binocular and monocular conditions may reflect a transfer of calibration, whereby participants adjusted their responses in the monocular condition based on prior experience in the binocular condition. The green shaded areas represent the acceleration phases for each condition.

Discussion

In this study, we compared stereoblind individuals and typically sighted controls on slant matching and reaching-to-grasp tasks to investigate how visuomotor function and stereo-motion perception develop without static stereopsis. Although slant perception was largely comparable across groups, the reaching-to-grasp task revealed important differences. As predicted, stereoblind individuals exhibited higher peak velocities and delayed emergence of grip orientation, suggesting a developmental adaptation that leverages alternative motion-based cues to support stereo-motion processing and guide actions.

Beyond Monocular Assumptions: Understanding Stereoblindness

A common intuition is that stereoblindness should yield performance broadly comparable to the control group's monocular viewing. Thus one would expect the stereoblind cohort to show similar patterns under binocular and monocular conditions.

However, our results do not support this assumption. The two groups converged under monocular viewing. Yet within the stereoblind group, binocular performance diverged from monocular performance, indicating that two-eye stimulation introduces signals that are not available monocularly despite the lack of usable static disparity. Thus stereoblindness does not preclude all binocularly derived information but merely reflects insensitivity to static binocular disparity, and when one eye is occluded both groups rely on the monocular cues in similar ways. Accordingly, comparable performance is expected and observed under monocular observation, consistent with prior work on slant perception in stereoblind observers using computer-generated stimuli.²⁶

Also, our results conflict with some previous studies. For example, Grant and Conway³⁴ found that individuals with impaired stereo vision exhibited slower peak velocities and longer movement times. A plausible explanation for this discrepancy concerns differences in kinematic measurement. In their study,³⁴ movement speed was derived from wrist velocity, which primarily indexes the transport component of the reach and may not fully capture digit adjustments during the reaching phase. In contrast, we quantified motion based on the centroid between the index finger and thumb, thereby incorporating local finger dynamics that can occur independently of wrist displacement. This methodological difference may reveal faster, more localized motion profiles that remain undetected in analyses restricted to wrist movement.

Stereo Motion and Reaching-to-Grasp

During reaching-to-grasp movements, individuals typically focus on the target rather than their hand,^35–38 perceiving binocular disparity of the hand and using disparity-matching to guide the trajectory.⁷ In stereoblind individuals, the absence of static disparity limits this strategy, but they may still rely on IOVD from their hand's optic flow to adjust grip orientation.

For observers with access to both CDOT and IOVD channels, an optimal transition between these mechanisms during reaching can be conceptualized as follows: CDOT, which operates more effectively at lower velocities, may contribute primarily to fine adjustments in the later phase of movement, whereas IOVD, optimized for higher velocities, is likely to dominate during the faster, mid-phase of the reach.

This theoretical framework may explain why stereoblind participants exhibited delayed grip inclination differentiation: it occurred closer to the peak velocity phase of movement so they could better perceive the alignment between their hand and the actual slant of the target. Consistent with this interpretation, Dogar et al.²⁵ reported a marked improvement in 3D motion discrimination as simulated MID speed increased from ∼6 to ∼100 cm/second in some of participants, comparable to the peak velocities in our stereoblind cohort. Although no significant Group × Slant interaction emerged in windows near peak velocity, an outcome plausibly attributable to inter-individual variability, the pattern is consistent with a late, speed-facilitated uptake of slant information for online control of hand orientation. An alternative interpretation is that the delayed GID in the stereoblind cohort reflects elevated between-subject variability, reducing early detectability; accordingly, this account remains provisional and motivates studies that parametrically manipulate approach speed and impose tighter task constraints (e.g., standardized start- and end-point positions and prescribed velocity profiles).

Notwithstanding these caveats, the hypothesis aligns with known acuity-dependent differences between CDOT and IOVD. CDOT is likely to depend on higher acuity to resolve finer positional disparities. In contrast, IOVD relies on velocity, favors lower spatial frequencies, and may even benefit from reduced acuity at higher retinal speeds.⁴⁰ This pattern is consistent with evidence that peripheral vision exhibits temporal sensitivity nearly comparable to foveal vision.⁴¹ When the eyes foveate the target, hand motion is primarily processed in the periphery, where stereoblind observers may retain IOVD signals supporting residual stereomotion sensitivity. Consistent with this view, Vater et al.⁴² demonstrated that peripheral vision remains highly functional for motion and action control, although the interaction between the two eyes’ peripheral fields is still poorly understood. The IOVD mechanism may provide a promising avenue to further investigate this binocular integration in peripheral vision. For people who can use CDOT, although it remains unclear how much visual acuity is required to detect changes in hand disparity during fixation, particularly for the use of CDOT in peripheral vision, individuals with superior stereovision may possess greater peripheral acuity, enabling finer detection of disparity changes. Alternatively, such individuals may effectively use both CDOT and IOVD mechanisms, with their peripheral vision relying less on high retinal velocities to extract MID information.

Regarding their perception of the static target's slant, in addition to contour and texture information, previous research suggests that non-binocular observers often utilize motion parallax by generating head movements along a frontoparallel plane.^43,44 It is plausible that stereoblind individuals in our study used a similar strategy. However, as we did not control or record head movements in the current experiment, this remains an open question for future research.

Motion Detection and Spatial Perception

Previous discussions of MID detection often focus on the mechanisms of detection itself,^{18,19,21,39,45} without addressing its functional significance. In practice, motion detection serves important perceptual purposes and guide action. For instance, motion parallax can be exploited monocularly to recover depth information about objects. With binocular vision, motion signals integrated across the two eyes can be utilized to support better depth perception and related spatial judgments. Empirical evidence indicates that in virtual environments lacking haptic feedback, proprioceptive input, or occlusion cues, the relative depth (i.e., allocentric distance) between a hand avatar and its target can be specified only with stereo vision, not monocularly.⁷ Thus the functional role of enhanced MID sensitivity, via CDOT, IOVD, or both, is to enhance discrimination of relative depth of objects. In reaching-to-grasp, stereomotion information supports more accurate alignment of grip orientation with target slant.

One might argue that hand orientation is governed primarily by proprioception rather than stereovision. This likely overstates proprioceptive reliability: studies showed that proprioceptive estimates are calibrated and visually recalibrated^46,47 and degrade near joint limits and with posture.^48–50 Thus stereomotion provides indispensable information for robust allocentric distance judgments and alignment of grip orientation beyond what proprioception can supply, which is a prediction open to empirical test.

Alternative Explanation for the Delayed GID

An alternative explanation for the delayed GID observed in the stereoblind group concerns the temporal dynamics by which monocular cues (e.g., contour, texture) and binocular cues (e.g., disparity, vergence) contribute to visuomotor control. Previous studies have shown that stereo cues were processed faster and influence visuomotor control earlier in the movement,^8,51 whereas monocular cues provide more accurate feedback during the later phase.⁵² This temporal dissociation can account for our findings: unlike the control group who could rely on stereo cues for a more planned trajectory of reaching-to-grasp, stereoblind individuals appeared to compensate by enhancing online interaction with texture and contour information during the final approach, maintaining grasping accuracy.

This framework can also account for the late Group × Slant interaction observed under binocular viewing. As shown in Figure 5 and Figure 6, control participants establish a more accurate slant to GI mapping early in the reach and therefore require only modest online adjustments thereafter. By contrast, stereoblind participants begin with a weaker slant specification; their initial hand orientation is less well aligned with the target and necessitates larger, distance-dependent corrections in the terminal segment, where close proximity permits comparison of hand orientation with the actual target slant.

However, this account is underspecified and functions more as an interpretive placeholder than a mechanistic theory. Notably, it cannot explain why stereoblind participants exhibited faster movement velocities only under binocular viewing; if the account were adequate, comparable dynamics should emerge under both monocular and binocular conditions. Because binocular estimates are constructed from monocular signals, the claim is not that binocular computation is faster per se, but that it attains criterion reliability earlier—owing to shorter temporal constants and reduced spatial-integration demands. Consequently, online control weights whichever cues are momentarily most reliable, with monocular cues gaining influence as evidence accumulates over time. This reframing sharpens the central questions: which binocular estimates drive the observed effect in Greenwald et al.,^8,51 and by what mechanisms do they achieve earlier reliability? Answering these questions likely requires targeted experiments that isolate stereomotion channels and other binocular cues, and that quantify their temporal dynamics and signal-to-noise ratios at task-relevant speeds.

Conclusions

Our findings reveal that, despite lacking static binocular disparity, stereoblind individuals exhibited efficient, albeit distinct, grasping behaviors compared to controls, indicating that the visuomotor system can recalibrate during development by seeking and utilizing available sensory cues. These results underscore the essential role of sensory experience in shaping eye-hand coordination and demonstrate that even when fundamental visual inputs are absent, stable motor control can emerge through flexible, adaptive developmental pathways. More importantly, they offer insight into how the visuomotor system ecologically adapts to changes in both external and internal environments by dynamically shifting between feedforward planning and real time information-based control.

Appendix.

Appendix A

Given the 3D coordinates of the thumb (x₁,y₁, z₁), and the index finger (x₂,y₂, z₂), we computed GI by first projecting both points onto the sagittal plane (i.e., where x = 0). The resulting projected points were:

$$\begin{array}{c}\hfill \mathrm{Point}\mathrm{A}:\left({x^{\text{'}}}_1,{y^{\text{'}}}_1,{z^{\text{'}}}_1\right)=\left(0,y_1,z_1\right)\end{array}$$

$$\begin{array}{c}\hfill \mathrm{Point}\mathrm{B}:\left({x^{\text{'}}}_2,{y^{\text{'}}}_2,{z^{\text{'}}}_2\right)=\left(0,y_2,z_2\right)\end{array}$$

We then defined a line connecting these two points on the sagittal plane and computed the angle between this line and the vertical (y) axis. GI was calculated as: GI = arctan ($\frac{z_2-z_1}{y_2-y_1}$). GI was defined such that a value of 0° corresponds to the index finger (Point B) being directly above the thumb (Point A) in vertical alignment, whereas 90° corresponds to the index finger being directly in front of the thumb in a horizontal configuration (See also Figure A1).

Figure A1.

(a) Schematic definition of GI—the angular orientation of the thumb–index aperture relative to the target surface. (b–d) Representative trial from the present study: (b) front view of digit trajectories during the reach-to-grasp, (c) right-lateral view of the same movement, and (d) the corresponding GI trajectory over the reach.

Appendix B

The results of HLM analyses are illustrated in Figure B1. In addition, fitted slopes of slant in different conditions are separately plotted in Figure B2 for readers to better compare them.

Figure B1.

Fixed-effect coefficients from HLM of GI fit in 50 contiguous distance windows over the last 40 cm of approach. The x-axis is absolute hand–target distance expressed as percent of the 40 cm interval; y-axis shows beta (β) estimates. Top: binocular; bottom: monocular. Curves show the estimated fixed effects for Group (blue), Slant (red), and Group × Slant (green). Filled symbols mark windows with P < 0.05. The Group effect emerged late under both viewing modes (≈78% of the trajectory for binocular; ≈94% for monocular) and persisted thereafter. The Slant effect appeared early (≈4% binocular; ≈12% monocular). A significant Group × Slant interaction occurred only late in the binocular condition (≈88%), indicating group-dependent slant sensitivity near contact; no interaction was detected under monocular viewing.

Figure B2.

Simple slopes (Slant) by group across absolute distance. Top: binocular; bottom: monocular. Lines show the fixed-effect estimate of the Slant coefficient (slope of Slant → GI) for Control and Stereoblind groups from HLM fit in 50 contiguous distance windows over the last 40 cm of approach (x-axis = % of 40 cm; y-axis = β estimate). Filled markers indicate windows with P < 0.05. Slant sensitivity is evident from the start in controls and emerges later in stereoblind observers, with convergence and a steady increase toward contact.

Appendix C

This figure shows GI over normalized movement time: control participants with binocular (top left) and monocular (top right) viewing, and stereoblind participants with binocular (bottom left) and monocular (bottom right) viewing. Each line represents one subject at one slant condition. In the control group, trajectories are tightly clustered with relatively low individual variability, indicating consistent visuomotor strategies and reliable depth estimation. In contrast, the stereoblind group exhibits substantially greater variability.

Figure C1.

Mean grip inclinations as a function of normalized movement time for each subject and each condition. Each line represent one subject at one slant condition.