ForcePinch: Force-Controlled XR Interaction
- ForcePinch is a spatial interaction technique that uses continuous pinch force to modulate pointer tracking speed in XR environments.
- It decouples gain control from hand movement by mapping pinch force to pointer speed, enabling both swift coarse positioning and careful fine adjustments.
- Empirical evaluations across 1D, 2D, and 3D tasks show that ForcePinch delivers high precision with trade-offs in operational efficiency and movement strategy.
ForcePinch is a force-responsive spatial interaction technique for XR that uses continuous pinch force as a secondary input channel to control pointer or object tracking speed during manipulation. While the user is already pinching to grab or drag an object, a lighter pinch yields faster movement for coarse positioning and a stronger pinch yields slower movement for precise adjustment. The method is framed through a “friction” metaphor in which more force means more friction and therefore less speed, and it is presented as a way to decouple gain control from hand distance or hand velocity, which are the control variables used by distance-responsive and speed-responsive techniques such as Go-Go and PRISM (Zhang et al., 24 Jul 2025).
1. Conceptual basis and interaction rationale
ForcePinch is motivated by the speed–precision trade-off in 3D interaction. In immersive environments, users often need to move rapidly across space and then transition to fine positioning near a target. Existing adaptive-gain techniques typically infer the desired gain from hand kinematics. Go-Go ties gain to hand displacement, and PRISM ties gain to hand velocity. ForcePinch instead makes gain depend on pinch force, which can be changed independently of hand distance or speed. This reassigns tracking-speed control from a kinematic variable to a force-responsive variable and thereby changes the structure of the interaction itself (Zhang et al., 24 Jul 2025).
The conceptual metaphor is grounded in everyday manipulation. The technique is explicitly inspired by the natural friction control inherent in the physical world, where fingertip force helps regulate both movement and stabilization. In ForcePinch, this embodied intuition is merged with the already-common pinch gesture used in XR for grabbing and selection. The method therefore does not introduce a separate modal command; it augments an existing gesture with a continuous control dimension. The paper frames this as a non-invasive embodied interaction in which a primary manipulation action and a secondary control signal coexist within the same gesture (Zhang et al., 24 Jul 2025).
A central design claim is that ForcePinch exposes desired gain more directly than motion-responsive techniques. In Go-Go and PRISM, users influence tracking speed only indirectly through how far or how fast they move. In ForcePinch, users can move quickly while intentionally slowing the pointer by squeezing harder, or keep the pinch light to preserve high speed without exaggerating reach or velocity. This decoupling is the main theoretical distinction of the method (Zhang et al., 24 Jul 2025).
2. Control model, transfer functions, and calibration
The runtime logic is described through three variables: pinch force , pointer tracking speed , and task-dependent baseline tracking speed . The mapping is negatively correlated: higher force produces lower tracking speed, and lower force produces higher tracking speed. The paper does not provide a single symbolic closed-form control equation for the full runtime update, but it does specify the transfer function through calibrated anchor points and interpolation (Zhang et al., 24 Jul 2025).
Each participant’s force range is normalized to , where $0$ is minimal calibrated force, $0.5$ is moderate calibrated force, and $1.0$ is maximal calibrated force. For ForcePinch, the three anchor points are: Thus normalized force $0$ maps to the fastest tracking speed $4c$, the moderate force point maps to the neutral speed 0, and maximal force maps to the slowest speed 1. The main paper describes cubic spline interpolation between these anchors, and the appendix specifies cubic Hermite spline interpolation. Out-of-range values are clamped to the nearest endpoint (Zhang et al., 24 Jul 2025).
The baseline 2 is task dependent. For the 1D and 2D tasks, 3. For the 3D task, 4. This yields a ForcePinch speed range from 5 down to 6 in 1D and 2D, and from 7 down to 8 in 3D (Zhang et al., 24 Jul 2025).
The comparison techniques are parameterized analogously. Constant uses a fixed C–D ratio 9. Go-Go keeps speed fixed at 0 for hand displacement below 1, then increases linearly to 2 at 3. PRISM uses 4 at 5 and increases linearly to 6 at 7 (Zhang et al., 24 Jul 2025).
| Technique | Control variable | Speed policy |
|---|---|---|
| Constant | none | fixed 8 |
| Go-Go | hand displacement | 9 below 0, then linear to 1 at 2 |
| PRISM | hand velocity | 3 at 4, linear to 5 at 6 |
| ForcePinch | pinch force | 7, 8, 9 with spline interpolation |
Calibration is individualized. Users apply three distinct pinch force levels, from minimum to maximum comfortable force, and each level is held for about one second while the system records a continuous force time series. The system then applies $0$0-means clustering with $0$1 to identify representative minimal, moderate, and maximal force levels. These become the user-specific landmarks for the mapping. The appendix states that cubic Hermite spline interpolation was preferred over linear interpolation because it provided a more uniform and natural control experience (Zhang et al., 24 Jul 2025).
3. Hardware prototype and system implementation
ForcePinch was implemented through a lightweight wireless sensing prototype built around a thin-film pressure sensor. The sensor was attached to the thumb pad and worn through an adjustable glove-and-ring arrangement. Pressure data were transmitted by an ESP32 microcontroller housed in a project box and carried in a waist-mounted bag. In the study, the sensing hardware remained attached across all four technique conditions so that hardware presence was held constant even when only ForcePinch consumed the force signal (Zhang et al., 24 Jul 2025).
The XR platform was a Meta Quest 3 running in mixed reality with hand tracking and passthrough enabled. The application was implemented in Unity using the Meta XR Interaction SDK. The authors extended the SDK’s built-in Distance Grab Interactable with a custom Movement Provider so that Constant, Go-Go, PRISM, and ForcePinch could all be realized within the same interaction framework. During the study, timestamps, object positions, hand positions, selection states, and current tracking speed were logged at $0$2 (Zhang et al., 24 Jul 2025).
ForcePinch included task-relevant feedback. A dynamic dot cursor at the pointer changed size with pinch force, and the cursor shrank as force increased. The intent was to make force magnitude visually legible through a “squeezing a soft object” metaphor. All techniques displayed tracking-speed-related cursor feedback, but ForcePinch additionally provided proprioceptive information through pinch-force modulation itself (Zhang et al., 24 Jul 2025).
The implementation also included a release-stabilization mechanism. When the user released the object, the object position reverted to the moment of peak force within the previous $0$3 seconds. This rollback mechanism was intended to reduce jitter caused by abrupt force drops or incidental hand motion at release, and it is one of the most concrete implementation details in the system design (Zhang et al., 24 Jul 2025).
4. Experimental evaluation across 1D, 2D, and 3D tasks
ForcePinch was evaluated in a within-subject study with $0$4 participants, including $0$5 female and $0$6 male participants aged $0$7–$0$8. VR experience varied from frequent use to no prior experience. Each participant completed three tasks—1D slider adjustment, 2D tracing, and 3D placement—under four techniques: Constant, Go-Go, PRISM, and ForcePinch. Technique order was counterbalanced by Latin square, and the study lasted about $0$9 minutes (Zhang et al., 24 Jul 2025).
The 1D slider task isolated endpoint control. ForcePinch achieved an error distance of $0.5$0 with CI $0.5$1, compared with $0.5$2 for PRISM, $0.5$3 for Constant, and $0.5$4 for Go-Go. ForcePinch and PRISM were statistically similar in error and both significantly better than Constant and Go-Go. ForcePinch’s operation time was $0.5$5 with CI $0.5$6, which did not differ significantly from the others, while the number of operations was $0.5$7 with CI $0.5$8, similar to PRISM and significantly better than Go-Go. On subjective measures, ForcePinch was less mentally demanding, less effortful, and less frustrating than Go-Go, and $0.5$9 of participants ranked ForcePinch first in this task (Zhang et al., 24 Jul 2025).
The 2D tracing task revealed ForcePinch’s strongest precision advantage and clearest efficiency cost. ForcePinch achieved the lowest tracing error distance at $1.0$0 with CI $1.0$1, outperforming Constant $1.0$2, PRISM $1.0$3, and Go-Go $1.0$4. However, it also produced the longest operation time, $1.0$5 with CI $1.0$6, versus $1.0$7 for Constant, $1.0$8 for Go-Go, and $1.0$9 for PRISM. The number of operations for ForcePinch was 0, comparable to PRISM and better than Go-Go, but worse than Constant. Hand distance was 1, significantly larger than all other techniques. Subjectively, ForcePinch was again better than Go-Go on all reported workload dimensions, but it was not clearly better than PRISM or Constant overall. In this task, ForcePinch is best characterized as the highest-precision but slowest technique (Zhang et al., 24 Jul 2025).
The 3D placement task reduced the distinctiveness of the force-responsive approach. ForcePinch achieved an error distance of 2 with CI 3, better than Go-Go 4 and Constant 5, but not clearly better than PRISM 6. Its operation time was 7 with CI 8, slower than PRISM 9 and Go-Go $0$0, and statistically similar to Constant $0$1. ForcePinch reduced the number of operations to $0$2, a $0$3 reduction relative to Constant, and reduced hand distance to $0$4, a $0$5 reduction relative to Constant, though both Go-Go and PRISM required less hand movement still. Subjective ratings showed no clear distinction among ForcePinch, Go-Go, and PRISM in 3D, while all three were better than Constant on several workload dimensions (Zhang et al., 24 Jul 2025).
| Task | Key ForcePinch metrics | Main comparative pattern |
|---|---|---|
| 1D slider | error $0$6, time $0$7, operations $0$8, hand distance $0$9 | precision comparable to PRISM and better than Constant and Go-Go |
| 2D tracing | error $4c$0, time $4c$1, operations $4c$2, hand distance $4c$3 | best precision, but slowest and largest hand movement |
| 3D placement | error $4c$4, time $4c$5, operations $4c$6, hand distance $4c$7 | better than Go-Go on error, weaker than PRISM on time |
5. Trade-offs, usage patterns, and design interpretation
The empirical profile of ForcePinch is not that of a uniformly dominant technique. Its strengths are concentrated in precision-heavy contexts, especially 1D endpoint adjustment and 2D tracing. Its weaknesses appear in efficiency, acceleration-heavy phases, and unconstrained 3D placement. The paper explicitly states that ForcePinch is best understood as a strong precision-oriented technique and as a broader proof of concept for force-responsive interaction in XR (Zhang et al., 24 Jul 2025).
One recurring finding is that ForcePinch appears better for deceleration than acceleration. The paper reports significantly more overshooting or movement than Constant in Tasks 1 and 3, and interprets this as a consequence of the inverse mapping: light force produces high speed, and high-speed phases therefore require low-force control while the hand is moving more. Near the goal, however, stronger force lowers gain and leaves more room for corrective motion. This supports an interpretation of ForcePinch as a braking or fine-adjustment mechanism rather than as a general acceleration mechanism (Zhang et al., 24 Jul 2025).
The study also reveals distinct user strategies. In 2D tracing, some participants adopted a sustained high-force precision mode, effectively turning ForcePinch into a slow, precise version of Constant. Others used dynamic modulation, keeping force lighter on straight segments and increasing force on curves or corners. This suggests that the technique is not a single behavior but a control space within which users can adopt different policies depending on task structure (Zhang et al., 24 Jul 2025).
The paper’s broader design reflection frames force-responsive interaction through four force-signal characteristics: magnitude, change, rhythm, and duration. In the implemented system, magnitude is the explicit control variable, but the authors treat the method as evidence that continuous force can function as an expressive secondary channel for embodied XR interaction more generally. This suggests a design space broader than speed control alone, although the reported evaluation remains focused on tracking-speed modulation (Zhang et al., 24 Jul 2025).
6. Position within pinch and force interaction research
ForcePinch addresses a limitation already visible in earlier XR pinch research: the standard thumb–index pinch is easy to implement and reliable because it depends on the distance between thumb and index finger, but it is also “not always natural,” can collide with other gestures, and can create sticky release behavior. A comparison of distance-based pinch, template-based grab, and controller interaction found that the controller was significantly faster and more accurate than both hand-based techniques, while Distance Pinch and Template Grab were not significantly different from each other in either speed or accuracy. That work suggests that a richer pinch vocabulary is needed and that binary geometric pinch alone is often overcommitted as an interaction primitive (Schäfer et al., 2022).
Within that context, ForcePinch can be read as one response to the overloading of binary pinch: it preserves the familiar pinch gesture but adds a continuous force channel. This does not solve all pinch-related problems, but it changes pinch from a purely geometric state detector into a composite interaction with separable grasp and gain-control semantics. A plausible implication is that force modulation can enrich pinch without requiring a wholly different hand pose or mode-switch gesture.
Research on around-device smartwatch input strengthens the same point from another direction. A sonar-based around-device system on an unmodified smartwatch found that a pinch-like thumb-to-index trigger detected from IMU data was feasible in isolation, with $4c$8 detection accuracy in pilot testing, yet it was slower, more error-prone, less comfortable, and more demanding than dwell and double-crossing in end-to-end selection. That result shows that recognition accuracy alone does not determine interaction quality. The timing and control consequences of a pinch-like action are equally important, which aligns with ForcePinch’s emphasis on continuous control rather than binary confirmation (Kim et al., 4 Mar 2025).
More recent sensing work suggests that dedicated pressure hardware is not the only possible substrate for force-responsive pinch interaction. Dorsal-hand deformation analysis in egocentric views improved index finger pinch estimation under self-occlusion and also supported binary detection of isometric “force clicks” with no discernible hand motion. Likewise, smartphone vibrometric force estimation showed that a phone’s vibration motor and IMU could estimate a pinch-type force with a mean absolute error of $4c$9 lbs on a profiled Pixel 4, although that method remained device-specific and closer to a pinch-gauge proxy than to standardized whole-hand grip measurement (Huang et al., 21 Jan 2026, Barry et al., 2 Dec 2025). These results suggest that future force-responsive XR systems may not be limited to dedicated thumb-mounted pressure sensors.
Taken together, ForcePinch occupies a distinct position in the literature. It is neither a conventional pinch recognizer nor a general force-estimation framework. It is a technique for using continuous pinch force as a control variable in XR manipulation. Its main contribution is to show that force can serve as a directly controlled secondary channel for tracking-speed modulation, yielding clear precision benefits in some contexts while exposing new trade-offs in efficiency, overshoot, and 3D movement strategy (Zhang et al., 24 Jul 2025).