Deployable Compliant Leg (DCL)

• The DCL is a reconfigurable leg module that deploys an elastic element to enable explosive, energy-efficient jumps without compromising normal locomotion.
• It employs a sector-shaped PEBA lattice with a Gyroid structure, validated via finite-element analysis and experimental tests on the Unitree Go2.
• Experimental results show a 17.1% increase in effective jump height when deployed, while the stowed state maintains near-baseline performance.

Deployable Compliant Leg (DCL) denotes, in the 2026 quadruped-jumping literature, a leg architecture in which a quadruped robot’s original leg is augmented with a bio-inspired elastic module that can be mechanically deployed for leaping and stowed for nominal locomotion. In the reported implementation on a Unitree Go2, the DCL acts as an additional elastic “spring leg” in parallel with the actuated joints during a squat–jump maneuver, storing energy during flexion and releasing it at takeoff; when stowed, it is kinematically disengaged and therefore does not add continuous parasitic stiffness to walking or trotting (Chen et al., 1 Mar 2026).

1. Concept, scope, and motivation

The DCL was introduced to address a specific limitation of electrically actuated quadrupeds: explosive vertical jumps require very high power over a short time interval, whereas motors remain bounded by peak torque and thermal limits, the torque–speed envelope and bus voltage, and the mass and cost penalties associated with actuator oversizing (Chen et al., 1 Mar 2026). The stated objective is therefore not generic whole-body compliance, but on-demand high-density elastic energy storage for maneuvers such as leaping.

Within this formulation, “deployable” has a precise mechanical meaning. The elastic element is physically switched into or out of the leg’s kinematic chain. In the deployed state it bridges the thigh and shank and behaves as a nonlinear spring and hard stop in the knee region; in the stowed state it folds along the side of the calf and is not in the load path. This distinction addresses a recurrent drawback of permanently integrated springs and parallel elastic actuators: parasitic stiffness during nominal gait, reduced joint control bandwidth, and increased cost of transport (Chen et al., 1 Mar 2026).

The design is explicitly bio-inspired. Its reference organism is the froghopper, whose semi-lunar process or pleural arch stores elastic energy during flexion and then releases it rapidly through a catapult-like mechanism. The DCL maps that principle onto a quadruped leg by combining a sector-shaped compliant element with a deployable flipping mechanism (Chen et al., 1 Mar 2026).

A common misconception is that the DCL is simply a permanently softer leg. The reported architecture is instead a reconfigurable leg assembly whose central claim is mode separation: high compliance for jump augmentation only when commanded, and near-baseline locomotion behavior when retracted. The stowed-state experiments, which showed only a $-0.4\%$ change in effective jump height relative to baseline, were presented precisely to support that distinction (Chen et al., 1 Mar 2026).

2. Bio-inspired operating principle

The biological analogue is the froghopper leg. In the cited mechanism, a curved sector-like arch around the trochanteral joint combines resilin, described as highly elastic and low hysteresis, with hard cuticle. During squatting or flexion, muscles slowly preload this structure; a latch holds the arch in the loaded state; and upon release the stored elastic energy is discharged rapidly, yielding power amplification beyond raw muscle power. The curved geometry also produces nonlinear stiffness: low stiffness at small deformation and higher stiffness as deformation grows (Chen et al., 1 Mar 2026).

The DCL reproduces this logic with engineering substitutions. A Sector-Shaped Compliant Module (SSCM) is mounted around the knee region; PEBA functions as the analog of resilin; and rigid PLA components serve as hard cuticle and attachment structures. During the squat phase, the thigh–shank angle closes to approximately $30^\circ\!-\!90^\circ$, compressing the sector-shaped PEBA lattice. Motor work is accumulated slowly as elastic energy in the lattice and in geometric bending. During takeoff, the joints extend and the SSCM releases its stored energy, augmenting actuator output according to the stated relation that total takeoff energy is the sum of motor energy and elastic energy (Chen et al., 1 Mar 2026).

The paper expresses the knee torque contribution of the compliant module as a function of knee angle $\theta$ by

$$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$

fit over the operating region $0^\circ<\theta\le 29^\circ$, with $R^2 \approx 0.87$ and a reported maximum torque of $\tau_{\max}\approx 6.8\,\text{N·m}$ in that range (Chen et al., 1 Mar 2026). The corresponding stored elastic energy is given by

$$E_{\text{elastic}}(\theta)=\int_0^\theta \tau_{\text{exo}}(\phi)\,d\phi =\frac{\alpha_3}{4}\theta^4+\frac{\alpha_2}{3}\theta^3+\frac{\alpha_1}{2}\theta^2+\alpha_0\theta.$$

Under the paper’s simplified vertical-jump model, this added elastic contribution increases takeoff energy and therefore center-of-mass jump height (Chen et al., 1 Mar 2026).

This operating principle places the DCL in a narrower category than continuously compliant legs or variable-stiffness joints. It is a catapult-like, event-specific energy buffer, not a continuously modulated impedance element. A plausible implication is that its strongest utility lies in impulsive behaviors—jumping, bounding, or other high-peak-power events—rather than in continuous cyclic force shaping.

3. Mechanical architecture and materials

The DCL consists of two tightly integrated subsystems: the SSCM and the deployable flipping mechanism. The SSCM uses Arkema Pebax® 3533 SP 01, a PEBA thermoplastic elastomer selected for high elasticity and flexibility, low hysteresis or low energy loss factor, excellent flex fatigue resistance, low density, and tunable hardness through the ratio of hard and soft segments. The rigid parts are PLA (Chen et al., 1 Mar 2026).

Its outer geometry is sector-shaped so that it occupies the otherwise unused angular volume between thigh and shank in a deep squat. This sector geometry serves two purposes stated in the paper: geometric compatibility with the knee region and progressive stiffness, because compression increases strain particularly near the inner radius and thus yields nonlinear stiffening (Chen et al., 1 Mar 2026).

Internally, the SSCM is filled with a Gyroid lattice, a triply periodic minimal surface generated in nTopology. The paper reports comparison among Diamond, Schwarz Primitive, Lidinoid, and Gyroid lattices, and selects Gyroid because of isotropic mechanical behavior, structural stability, smooth continuous surfaces with lower stress concentration than node-based lattices, and self-supporting geometry in 3D printing. The Gyroid is described by

$\sin x \cos y + \sin y \cos z + \sin z \cos x = t.$

The resulting module is therefore a PEBA shell shaped as a sector and filled with a Gyroid lattice that acts as the tendon-like energy storage medium (Chen et al., 1 Mar 2026).

The flipping mechanism realizes the deployable functionality. In the stowed state, the module is folded along the side of the calf and removed from the load path. In the deployed state, it rotates by approximately $90^\circ$ to align with the thigh and engage the knee region. The reported components are the SSCM lattice, push rod, top cap, rotating sleeve, return spring, inner fixed post, guide pin, sliding block, and locking pin. Linear actuation of the push rod moves the sliding block along the vertical axis; a guide pin follows a helical cam track around the rotating sleeve; and the translation is converted into a precise $30^\circ\!-\!90^\circ$0 rotation. The return spring maintains preload so that the locking pin snaps into detents in both end states, yielding bistable behavior with no continuous holding torque required in either state (Chen et al., 1 Mar 2026).

Mechanically, this is the defining distinction of the DCL relative to ordinary parallel springs. It is not only compliant; it is reconfigurable. This suggests an architectural niche in which geometric switching, rather than continuous impedance modulation, is used to control when elasticity is present in the leg.

4. Analysis, finite-element modeling, and operating limits

The SSCM was analyzed in Abaqus/Explicit using a hyperelastic Marlow material model for PEBA, based on Arkema’s uniaxial test data. The reported Poisson ratio is $30^\circ\!-\!90^\circ$1, chosen to represent near incompressibility while avoiding locking, and the mesh uses C3D10M modified quadratic tetrahedral elements for large deformation, contact, and mesh robustness (Chen et al., 1 Mar 2026).

The finite-element analysis had two stated purposes: validating structural integrity under large deformation and extracting torque–angle characteristics for control and performance estimation. A kinematic coupling at the knee center of rotation was used to replicate the Unitree Go2 knee joint, and the knee angle $30^\circ\!-\!90^\circ$2 was swept from $30^\circ\!-\!90^\circ$3 to $30^\circ\!-\!90^\circ$4. Within that sweep, the paper distinguishes three regimes: an operating range of approximately $30^\circ\!-\!90^\circ$5, a safety margin from $30^\circ\!-\!90^\circ$6 to $30^\circ\!-\!90^\circ$7, and a densification region beyond approximately $30^\circ\!-\!90^\circ$8, where lattice collapse causes a sharp rise in stiffness (Chen et al., 1 Mar 2026).

These limits are central to the design. The DCL is not meant to exploit unrestricted large deformation. Instead, it uses a deliberately bounded compliant regime for energy storage, followed by a sharply stiffening region that acts as a hard safety limit against catastrophic collapse. This is a nonlinear protective feature rather than a side effect.

The paper’s simplified energetic interpretation connects the FEA model to jump performance. If the SSCM contributes an additional elastic energy increment $30^\circ\!-\!90^\circ$9, the change in jump height is approximated by

$\theta$0

The reported $\theta$1 increase in vertical jump height is then interpreted as evidence of a significant additional energy contribution from the SSCM at essentially the same squat depth and battery state (Chen et al., 1 Mar 2026).

A plausible implication is that DCL design is governed by a three-way balance among lattice density, shell geometry, and actuator preload capacity. Structures that are too soft risk bottoming out or entering densification too early; structures that are too stiff may exceed the motor’s capacity to preload them effectively. The paper explicitly states that this balance was tuned through FEA-based design of lattice density and shell geometry (Chen et al., 1 Mar 2026).

5. Integration with Unitree Go2 and experimental results

The experimental platform is the Unitree Go2 quadruped. The SSCM is mounted externally near the knee region of each leg. When deployed, it bridges the thigh and shank so that knee flexion compresses the module; when stowed, it folds alongside the calf and the leg behaves like the original unit. The robot retains its original actuators; the DCL adds passive structures and the flipping mechanism rather than replacing the existing drive system (Chen et al., 1 Mar 2026).

The jump protocol was tightly controlled. All experiments used a standardized squat height

$\theta$2

battery state of charge was kept above $\theta$3, each configuration was tested for $\theta$4 trials, and reported values are means. Effective jump height was defined as

$\theta$5

and the relative change with respect to baseline was defined as

$\theta$6

Motion capture used a Luster FZMotion system with approximately $\theta$7 mm positional accuracy and 15 reflective markers: 3 on the trunk and 3 on each thigh, so that each segment could be tracked as a rigid body with full 6-DoF motion (Chen et al., 1 Mar 2026).

Three conditions were compared: baseline with no DCL, stowed with the SSCM installed but disengaged, and deployed with the SSCM engaged. The reported results were as follows.

Configuration	Effective jump height $\theta$8	Change vs. baseline
Baseline	373.1 mm	N/A
Stowed	371.7 mm	$\theta$9
Deployed	437.1 mm	$$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$0

The corresponding peak heights were $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$1 mm for baseline, $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$2 mm for stowed, and $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$3 mm for deployed (Chen et al., 1 Mar 2026).

These measurements support two distinct claims. First, the deployable architecture largely preserves baseline performance when the module is not engaged: the stowed condition changed effective jump height by only $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$4. Second, when engaged, the module increased effective vertical jump height from $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$5 mm to $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$6 mm, i.e. by $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$7 mm or $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$8, at the same squat depth (Chen et al., 1 Mar 2026).

The paper is more circumscribed about control than about mechanics. It states that the stiffness model is intended for integration into feedforward torque or segmented PD control, while future work may use MPC or reinforcement learning to exploit the nonlinear compliance more fully (Chen et al., 1 Mar 2026). It does not claim a fully developed compliance-aware controller in the current prototype.

6. Trade-offs, neighboring research directions, and future development

The paper identifies several explicit trade-offs. One is stiffness versus energy storage: softer structures permit larger deformation and larger energy-storage stroke but risk bottoming out, whereas stiffer structures store energy at smaller deformation but are harder to preload with limited motor torque. Another is mass versus benefit: the compliant structure must store meaningful energy while keeping mass low. A third is deployability versus complexity: the flipping mechanism introduces additional parts and potential failure points, although the reported design remains cam-based and bistable (Chen et al., 1 Mar 2026).

Its limitations are equally explicit. Long-term fatigue testing under robotic duty cycles is not yet reported. The SSCM has a fixed passive stiffness profile and cannot be tuned online like a full variable-stiffness actuator. The study focuses on vertical jump height rather than landing behavior. Automatic deployment is not yet implemented; the present prototype uses manual triggering of the flipping mechanism (Chen et al., 1 Mar 2026).

In the broader compliant-leg literature, the DCL occupies a specific position. Work on the Spring-Loaded Inverted Pendulum (SLIP) model treats compliant legs as hybrid dynamical templates and shows how passive dynamics, finite stability, and viability can structure gait control with minimal inputs (Salazar et al., 2011). Research on hybrid passive–active compliance shows that distributing stiffness between physical springs and virtual control can preserve landing robustness under large sensorimotor delays or low update frequencies (Ashtiani et al., 2021). Studies of terrain-adaptive stiffness modulation in Dual-SLIP locomotion show that leg stiffness is a primary control variable when the environment becomes compliant (Karakasis et al., 2022). These works do not propose the DCL mechanism itself, but they provide analytical and control frameworks into which an on-demand spring leg could plausibly be embedded.

Neighboring mechanical paradigms differ in where they place compliance. Tensegrity-based legs modulate stiffness by changing cable pre-tension in a structurally compliant joint network (Mortensen et al., 28 Apr 2025). Fibre-jammed tendons vary both stiffness and damping through vacuum-driven interfibre friction (Liow et al., 2023). Continuously compliant lower legs replace rigid shanks and discrete series springs with deformable structural members controlled through end-point force regulation (Bendfeld et al., 2024). Miniature origami systems such as CLARI emphasize modular deployability and body-shape adaptation through flexures and laminate mechanisms (Kabutz et al., 2023). Reconfigurable five-bar legs alter geometry to switch between height-advantaged and force-advantaged configurations, though without the intrinsic compliant catapult principle of the DCL (Harris et al., 13 Nov 2025). Relative to these lines, the DCL is best understood as an architecture for mechanically deployable, event-triggered parallel elasticity rather than continuously engaged compliance or continuously tunable stiffness.

The future directions named in the DCL paper follow from that positioning. The concept is said to be extendable to bipeds, hexapods, or even arms that occasionally require high-power motions. Suggested directions include automatic deployment through passive latching or a micro-actuator, multi-objective optimization of gyroid cell size, wall thickness, and sector geometry, graded or variable-density lattices, integration with variable-stiffness concepts, and more advanced controllers such as MPC or reinforcement learning (Chen et al., 1 Mar 2026).

The central technical takeaway is therefore narrow but substantive. In the reported meaning of the term, a Deployable Compliant Leg is not merely a compliant leg and not merely a reconfigurable leg. It is a quadruped leg augmented with a sector-shaped PEBA lattice module whose elasticity is mechanically inserted into the leg only when explosive power amplification is needed, and removed when nominal locomotion should remain near-rigid. On the reported Unitree Go2 implementation, that architecture produced a $$\tau_{\text{exo}}(\theta)=\alpha_3\theta^3+\alpha_2\theta^2+\alpha_1\theta+\alpha_0,$$9 increase in effective vertical jump height while the stowed state remained near baseline (Chen et al., 1 Mar 2026).