HapticGiant: Large-Scale Kinesthetic Haptic Interface

• HapticGiant is a large kinesthetic haptic interface that features an 8-DOF design combining a 2D gantry and a 6-DOF manipulator to emulate human arm movements.
• It employs a novel admittance-type force control using Hierarchical Quadratic Programming to manage joint, Cartesian constraints, and singularities in real time.
• Experimental results demonstrate robust force/torque rendering, expansive workspaces for natural locomotion, and high fidelity in virtual mechanics simulation.

HapticGiant is a very large kinesthetic haptic interface introduced as a grounded system for immersive virtual reality that aims to match the properties of the human arm as closely as possible while permitting natural user locomotion and full haptic feedback (Fennel et al., 13 Aug 2025). Its defining characteristics are an 8-DOF architecture composed of a 2D gantry and a custom 6-DOF overhead manipulator, and a novel admittance-type force control scheme based on hierarchical optimization that renders arbitrary serial kinematic chains and Cartesian admittances while natively accounting for joint constraints, Cartesian constraints, and singularities (Fennel et al., 13 Aug 2025).

1. System definition and architectural composition

HapticGiant was presented in "HapticGiant: A Novel Very Large Kinesthetic Haptic Interface with Hierarchical Force Control" (Fennel et al., 13 Aug 2025). The stated motivation is that kinesthetic haptic interfaces still face limited workspaces, insufficient degrees of freedom, and kinematics that do not match the human arm, even as head-mounted displays have become commercially available (Fennel et al., 13 Aug 2025).

The hardware system comprises a 2D gantry, termed the "Prepositioning Unit" or PPU, and a custom 6-DOF overhead manipulator. The PPU provides large planar movement for unconstrained user locomotion over $4.7 \times 5.5$ meters, while the manipulator is dimensioned and optimized to match the human arm. The manipulator uses six Sensodrive Sensojoint series elastic actuators and carbon fiber links, includes a slip ring at Joint 1 for infinite rotation, and carries a 6-DOF force/torque sensor at the end-effector. Fast, low-latency communication is provided by EtherCAT. A physical prototype of this 8-DOF robot was constructed and tested (Fennel et al., 13 Aug 2025).

Subsystem	Reported role	Reported detail
2D gantry (PPU)	Large planar movement	$4.7 \times 5.5$ meters
6-DOF overhead manipulator	Human-arm-matched haptic rendering	Six Sensodrive Sensojoint actuators, carbon fiber links
Slip ring at Joint 1	Continuous rotation	Infinite rotation
End-effector sensor	Interaction measurement	6-DOF force/torque sensor
Communication	Low-latency control	EtherCAT

This architecture places HapticGiant in the class of grounded kinesthetic interfaces rather than tactile displays or wearable cueing systems. A plausible implication is that its contribution is not miniaturization or portability, but the coupling of human-scale workspace, arm-compatible kinematics, and force-feedback authority within a single system.

2. Design goals: workspace, degrees of freedom, and kinematic compatibility

The paper identifies several explicit design goals: a large workspace, a full 6-DOF end-effector, kinematic compatibility with the human arm, high force/torque capacity, functional safety, and extensibility and reproducibility through off-the-shelf components wherever possible (Fennel et al., 13 Aug 2025). The manipulator’s joint architecture is described as specifically optimized through anthropometry and genetic algorithms to match the human arm’s pose, velocity, and acceleration spaces (Fennel et al., 13 Aug 2025).

The workspace emphasis differentiates HapticGiant from earlier grounded interfaces designed for desktop or CAD contexts. For example, eM-Virtual Desktop integrated Phantom Desktop devices that provided 6 positional degree-of-freedom sensing, 3 degrees of force feedback, a work envelope of $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$, and a maximum force of $6.4$ N, with separate graphics, collision-detection, and haptic loops (0705.1285). HapticGiant instead targets room-scale interaction and walkable use, with the explicit goal of enabling users to embody virtual avatars or manipulate digital twins of complex mechanisms such as an excavator or a door (Fennel et al., 13 Aug 2025).

Within the broader haptics landscape represented by the provided literature, this problem framing is distinct. FlyHaptics addresses drone-mounted multi-contact tactile feedback in mid-air (Moreno et al., 5 May 2025); the soft hydro-pneumatic haptic ring addresses roughness, thermal, and softness cues on the proximal phalanx while leaving the fingertip unobstructed (Cozcolluela et al., 23 Mar 2025); ArrayTac addresses simultaneous rendering of shape, stiffness, and friction through a $4 \times 4$ piezoelectric tactile display (Liang et al., 14 Mar 2026). HapticGiant instead targets whole-arm kinesthetic interaction under grounded mechanical constraints (Fennel et al., 13 Aug 2025).

A common misconception is that all haptic interfaces solve the same rendering problem. The cited systems indicate otherwise: HapticGiant is concerned with kinesthetic force interaction over a large reachable volume, whereas tactile displays and wearables focus on local cutaneous rendering, and aerial systems focus on ungrounded contact delivery.

3. Hierarchical admittance control

The central technical contribution is a novel admittance-type force control algorithm built on Hierarchical Quadratic Programming, or HQP (Fennel et al., 13 Aug 2025). The paper states that the controller renders both arbitrary serial kinematic chains and Cartesian admittances, while natively handling joint limits, workspace limits, singularity avoidance, and kinematic coupling between the physical manipulator and the virtual model (Fennel et al., 13 Aug 2025).

The virtual system is modeled in joint space as

$$\mathbf{M}^{a}\ddot{\mathbf{q}}^{a} + \mathbf{c}^{a}(\mathbf{q}^{a}, \dot{\mathbf{q}}^{a}) = \mathbf{J}^{aT}\left(\mathbf{w}^{B}_{E} - \mathbf{w}^{B}_{E,\mathrm{ref}}\right) - \boldsymbol{\tau}^{a}_{\mathrm{dis}} + \boldsymbol{\tau}^{a}_{\mathrm{dri}},$$

where the quantities correspond to the inertia matrix of the digital twin, nonlinear dynamics, Jacobian, user-applied wrench, reference wrench, and dissipative and driving torques (Fennel et al., 13 Aug 2025). The controller solves for behavior that renders the virtual system subject to all active constraints.

The HQP formulation is described as task-prioritized. The reported task stack includes, in order of priority, kinematic coupling; manipulator joint angle and velocity limits; singularity avoidance; Cartesian workspace limits for the end-effector; Cartesian limits for the elbow; digital-twin joint angle and velocity limits; acceleration limits for manipulator and digital twin; admittance dynamics; constraint torque ranges; and constraint torque minimization (Fennel et al., 13 Aug 2025). The stated purpose of the recursive nullspace construction is that lower-priority tasks cannot interfere with more critical constraints such as safety (Fennel et al., 13 Aug 2025).

The paper further states that singularity avoidance is implemented by treating joint combinations as virtual joints and constraining them away from $0$ and $\pi$, and that constraint torques are direction-limited so that only barrier-like responses are produced in the direction opposing the violated limit (Fennel et al., 13 Aug 2025). This suggests that the controller does not treat safety or feasibility as after-the-fact corrections; instead, they are embedded directly in the optimization problem.

4. Native treatment of constraints and real-time execution

A recurring theme in the description of HapticGiant is that system limitations are incorporated natively rather than appended as post-processing (Fennel et al., 13 Aug 2025). Joint and velocity limits for both the physical robot and the digital twin are bounded using analytical formulas for discrete-time acceleration integration under constraints. Cartesian limits are enforced not only at the end-effector but also at intermediate links such as the elbow, using Jacobian-based mappings from control variables to constrained Cartesian motion. When a limit is reached, the generated constraint torques are fed back physically into both the manipulator and the virtual admittance, preserving consistent interaction (Fennel et al., 13 Aug 2025).

The reported implementation is in C++, running at $1$ kHz. The average controller compute time is $529\,\mu\text{s}$, and the $4.7 \times 5.5$0th percentile is $4.7 \times 5.5$1 for the 8-DOF system under the stated constraints (Fennel et al., 13 Aug 2025). For a kinesthetic haptic interface, these numbers matter because the control loop must remain stable while solving a constrained optimization problem at haptic rates.

The comparison with earlier haptic software frameworks is instructive. eM-Virtual Desktop also used a $4.7 \times 5.5$2 Hz haptic loop, but did so in a client-server architecture that separated graphics, collision detection, and haptic feedback into three loops, with the Phantom Desktop providing force but not torque (0705.1285). HapticGiant’s contribution is different: it places a hierarchical constrained optimizer inside the primary force-control loop of a full-scale physical mechanism (Fennel et al., 13 Aug 2025). A plausible implication is that the difficulty lies not only in achieving haptic-rate control, but in doing so while simultaneously respecting large-robot kinematics, singularities, and user safety.

5. Experimental characterization and demonstrated capabilities

The paper reports a physical workspace of approximately $4.7 \times 5.5$3, end-effector force and torque capabilities of up to $4.7 \times 5.5$4 N and $4.7 \times 5.5$5 Nm in any direction, and infinite yaw at the base enabled by the custom slip ring (Fennel et al., 13 Aug 2025). Using the Haptify metrics, the reported comparison to commercial table-top devices includes a workspace volume $4.7 \times 5.5$6 larger, mean free-space force rendering of approximately $4.7 \times 5.5$7 N, stiffness rendering accuracy up to $4.7 \times 5.5$8 of the intended value, and stiffness peak force of $4.7 \times 5.5$9 N. The paper also notes that free-space vibrations are greater than for small table-top devices, which limits very fine texture rendering (Fennel et al., 13 Aug 2025).

The demonstrations include 6-DOF Cartesian admittance rendering in which users can walk and move freely while the system enforces mechanical constraints, workspace boundaries, and singularities with dynamically consistent feedback. The system is also reported to render digital-twin tasks involving complex serial kinematic chains such as excavator arms or doors with co-dependent locks and latches (Fennel et al., 13 Aug 2025). User-study results based on NASA TLX questionnaires with $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$0 are described as showing low physical and mental workload for standard tasks, with the "door" scenario median in the lowest third (Fennel et al., 13 Aug 2025).

These results support two distinct claims. First, HapticGiant is not only a large robot with force feedback; it is a haptic system designed to render constrained virtual mechanics. Second, enlargement of workspace does not automatically preclude high stiffness rendering, although the reported increase in free-space vibrations indicates a tradeoff. A plausible summary is that the interface appears optimized for large-scale kinesthetic realism rather than for the finest cutaneous texture cues.

6. Position within the wider haptics ecosystem

HapticGiant occupies one region of a broader research space in which different systems emphasize different sensory channels, workspaces, and interaction paradigms. The relation is best understood by contrast rather than by subsumption.

In teleoperation, "Integrating and Evaluating Visuo-tactile Sensing with Haptic Feedback for Teleoperated Robot Manipulation" equipped a Tiago robot with GelSight Mini sensors and provided haptic feedback to MANUS Prime X Haptic gloves; in a user study with seven participants, haptic feedback reduced average object deformation by $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$1 during manipulation of a soft object (Becker et al., 2024). This illustrates the continued importance of force-related feedback in remote manipulation, a domain that the HapticGiant paper identifies as relevant through virtual reality and teleoperation scenarios (Fennel et al., 13 Aug 2025).

In tactile content creation, "Controllable Visual-Tactile Synthesis" generated both RGB imagery and dense tactile gradient maps from sketches for rendering on an electroadhesion-based haptic surface (Gao et al., 2023), while "HapticMatch" introduced a precisely aligned dataset of RGB images, surface height maps, and friction-induced vibrations for $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$2 diverse materials and proposed a "Scan-to-Touch" workflow using diffusion and flow-matching models (Zhang et al., 23 Jan 2026). These systems focus on tactile asset synthesis rather than large-scale kinesthetic embodiment. A plausible implication is that such content-generation pipelines could complement, rather than replace, kinesthetic platforms by supplying richer tactile semantics for multimodal VR environments.

In tactile and wearable hardware, ArrayTac demonstrated simultaneous rendering of shape, stiffness, and friction, including a tele-palpation experiment over more than $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$3 km with $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$4 accuracy in identifying the number and type of tumors in a breast phantom (Liang et al., 14 Mar 2026). The soft hydro-pneumatic haptic ring rendered roughness, thermal, and softness cues on the proximal phalanx, with participants achieving up to $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$5 accuracy in texture matching (Cozcolluela et al., 23 Mar 2025). FlyHaptics demonstrated a drone-mounted multi-contact aerial interface with mean pattern-recognition accuracy of $$16 \text{ cm} \times 13 \text{ cm} \times 13 \text{ cm}$$6 in a grounded pilot study and stable hover performance during flight demonstrations (Moreno et al., 5 May 2025). These works show that contemporary haptics research is increasingly multimodal and application-specific. HapticGiant’s specific role is the large-scale kinesthetic end of that spectrum.

From this perspective, HapticGiant is best characterized not as a general replacement for tactile, wearable, or aerial systems, but as a platform for human-scale force interaction where locomotion, arm-compatible kinematics, constraint-consistent dynamics, and high force authority are primary requirements (Fennel et al., 13 Aug 2025).