Husky Carbon: Legged-Aerial Hybrid Robot

• Husky Carbon is a hybrid quadrupedal robot that integrates legged locomotion with thruster-assisted aerial mobility to address multimodal challenges.
• It employs a morpho-functional design driven by Mobility Value of Added Mass (MVAM) to optimize forward walking efficiency and payload allocation.
• The platform features advanced actuator, sensing, and control systems, enabling tethered trotting and WAIR-inspired thruster-assisted incline locomotion in simulation.

Searching arXiv for Husky Carbon and closely related papers. Husky Carbon is a lightweight quadrupedal robot under development as a legged–aerial multi-modal platform that combines legs for terrestrial locomotion with thrusters for aerial or partially aerial mobility. In the Husky Carbon literature, “multi-modal locomotion” denotes the integration of grounded legged locomotion, thrust-assisted locomotion, and intended legged-to-aerial and aerial-to-legged transitions within a single machine rather than the addition of propellers as a mere stabilization aid. The platform is motivated by bird locomotion, especially the coexistence of intermittent ground contact and aerial force generation, and by field-robotics use cases such as search and rescue, inspection, and reconnaissance in cluttered, damaged, or unstructured environments (Salagame et al., 2022).

1. Definition, development trajectory, and problem setting

The Husky Carbon project is explicitly framed as an exploratory robotics program rather than a completed flying quadruped. The 2021 design paper presents Husky Carbon as a morpho-functional hybrid robot intended to combine quadrupedal legged locomotion and a future aerial/rotorcraft mode within a single platform, while the 2022 progress report emphasizes hardware integration, simulation infrastructure, and demonstrated locomotion behaviors rather than an end-state multimodal system (Ramezani et al., 2021, Salagame et al., 2022). The 2023 thesis and subsequent incline-walking papers extend the program toward trotting-hovering and thruster-assisted incline walking, especially through simulation of WAIR-inspired behaviors (Wang, 2023, Salagame et al., 2023).

Stage	Paper	Reported emphasis
2021	"Generative Design of NU's Husky Carbon, A Morpho-Functional, Legged Robot" (Ramezani et al., 2021)	MVAM, front-heavy morphology, additive manufacturing, preliminary untethered trotting, gait simulation
2022	"A Letter on Progress Made on Husky Carbon: A Legged-Aerial, Multi-modal Platform" (Salagame et al., 2022)	Integrated custom hardware, sensing and control stack, multimodal planning simulation, tethered trotting
2023–2024	"LeggedWalking on Inclined Surfaces" (Wang, 2023), "Quadrupedal Locomotion Control On Inclined Surfaces Using Collocation Method" (Salagame et al., 2023), "Thruster-Assisted Incline Walking" (Krishnamurthy et al., 2024)	WAIR-inspired modeling and thrust-assisted steep-slope locomotion

The central research problem is the coexistence of antagonistic design constraints. Aerial mobility favors low mass, low drag, and tight power budgets; legged mobility favors structural robustness, support-polygon margin, and actuation against ground reaction forces. The Husky Carbon papers repeatedly treat this not as a secondary implementation difficulty but as the defining systems problem of the platform. The resulting machine is therefore best understood as a testbed for morphology, actuation, control, and planning under conflicting terrestrial and aerial requirements rather than as a narrowly optimized quadruped or a conventional multirotor (Salagame et al., 2022).

2. Morpho-functional design and the Mobility Value of Added Mass

The 2021 design study formulates Husky Carbon through the Mobility Value of Added Mass (MVAM) problem, a morphology-driven mass-allocation problem restricted to the relation between morphology and energetic efficiency of legged mobility (Ramezani et al., 2021). MVAM asks, in effect, how mass should be distributed so that Total Cost Of Transport (TCOT) is minimized while preserving payload margin for future aerial hardware. The key descriptors used in the MATLAB evaluation are the fore–aft and vertical COM positions, $C_x$ and $C_y$, together with the sagittal-plane mass moment of inertia $I_b$.

The generative-design workflow couples Grasshopper inside Rhino 6, Galapagos as the evolutionary solver, and a MATLAB dynamics evaluation. The parametric design space includes the location of harmonic drive components, electronics, housing sizes, body-frame dimensions, and connecting-rod geometry. Candidate morphologies are then evaluated under comparable predefined forward walking gaits rather than co-optimized gait–morphology pairs. The authors therefore identify morphology-to-efficiency trends, not a full-body optimal-control solution.

The main reported result is that front-heavy morphologies outperform back-heavy ones for the constrained forward walking task. The authors state that in forward walking “hosting larger added mass in a front heavy morphology is less costly,” that such morphologies achieve lower TCOT, and that they retain larger allowable margins on sagittal-plane inertia $I_b$ (Ramezani et al., 2021). In the simulation reported for the selected front-heavy morphology, the authors give a forward walking speed of 0.2 m/s, gait period of 0.25 s, and TCOT of 0.2. They also note an important limitation: because the gait was predefined, the front-heavy conclusion is task- and gait-dependent.

A plausible implication is that Husky Carbon’s morphology was never intended to minimize mass in the abstract. Rather, the project treats morphology as a reserve-allocation problem: every gram assigned to structure, actuation, or sensing competes with future multimodal capability. That systems perspective is one of the project’s most distinctive features.

3. Embodied architecture: structure, actuation, sensing, and power

At the platform level, Husky Carbon is a quadruped with four legs, each having 3 degrees of freedom, for a total of 12 actuated joints. The legs are described as two pairs of identical parallelogram mechanisms. The body is fabricated primarily from reinforced thermoplastic materials using additive manufacturing, with a major design theme being the embedding of transmissions and actuator components directly into printed structural elements to reduce separate metal housings and fasteners (Salagame et al., 2022).

Across the reports, the platform dimensions reflect different embodiments or reporting conventions. The 2021 design paper describes the implemented robot as 0.8 m tall, 0.3 m wide, and 4.3 kg in mass; the 2022 progress report describes the prototype as about 0.4 m tall in quadrupedal posture, about 0.3 m wide, and 4.3 kg; and the incline-locomotion work later reports a 4.3 kg legged platform plus a 3.3 kg propulsion unit for a total mass of 7.6 kg in the thruster-equipped configuration (Ramezani et al., 2021, Salagame et al., 2022, Salagame et al., 2023).

Each of the 12 joint actuators is built around a T-Motor brushless winding with 400 KV, with transmissions based on harmonic drive component sets. The explicitly reported gear ratios are 30:1 for the knee, 50:1 for the hip sagittal joint, and 100:1 for the hip frontal joint. The progress report’s system figure names the motors as T-Motor MN4004 400KV and the servo drives as Elmo Gold Solo Twitter, while the thesis later describes T-motor Antigravity 4006 brushless motors, Harmonic Drive CSF-11-30-2A-R, ELMO Gold Twitter servo drives, and RLS RMB20 encoders. The aerial subsystem is described in multiple iterations: a conceptual 2021 design in which the legs move sideways and the knee actuator, through a clutch mechanism, would drive a propeller; a 2022 top-mounted four-thruster arrangement; and later thruster-equipped configurations using four ducted fans and, in the thesis, DS-30-AXI HDS electric ducted fans (69 mm) with YGE 95A LV ESCs (Salagame et al., 2022, Wang, 2023).

The sensing and compute stack also evolves over time. The early design paper states that the prototype lacks exteroceptive sensors and relies on off-board body orientation sensing. The progress report later specifies a VectorNav VN-100 IMU, an Intel RealSense T265 tracking camera, hall-effect encoder sensing at the joints, and a Speedgoat IO581 Real-Time Target Machine running Simulink Real-Time and communicating with servo drives over EtherCAT. The thesis then adds a higher-level architecture with a ROS controller on an NVIDIA Jetson Nano and a Pixhawk 2.4.8 flight controller linked via MAVLink (Ramezani et al., 2021, Salagame et al., 2022, Wang, 2023).

A persistent practical constraint is power. Both the 2021 and 2022 papers state that the prototype currently operates using an external power supply, and the 2022 test arena includes a large fan for actuator cooling. Untethered onboard power is therefore not established in the core progress report (Ramezani et al., 2021, Salagame et al., 2022).

4. Reduced-order modeling, multimodal planning, and legged control abstractions

The 2022 progress report formalizes Husky Carbon’s reduced-order kinematics by assigning each leg $i \in \{1,2,3,4\}$ the coordinates hip frontal angle $(\phi_i)$, hip sagittal angle $(\psi_i)$, and leg length $(l_i)$. The resulting reduced-order model comprises 12 kinematic DOFs and 6 dynamical DOFs, with the robot reduced to a single rigid body with 6-DOF dynamics and massless leg linkages; the authors state that the dynamics can be derived using the Euler–Lagrangian formulation (Salagame et al., 2022).

This reduced-order model underpins path planning and trajectory tracking. The reported planning environment is discretized in two ways for comparison: uniform discretization into points and a 3D MM-PRM, i.e. a multimodal probabilistic roadmap adapted to mode switching. The reported simulation shows successful navigation together with explicit legged-to-aerial and aerial-to-legged transformation sequences. The same section notes that, in that simulation, propellers were attached near the leg end, which indicates that propulsion placement remained an active design variable rather than a fixed architectural decision.

On the control side, the progress report gives a partial but informative view. For legged locomotion it mentions an RG-based framework that manipulates kinematic references to satisfy the friction pyramid constraint, thereby avoiding heavier optimization-based constraint handling at the high level. The paper also reports effective tracking of predefined joint trajectories and body-pose stabilization using camera and inertial feedback during trotting-in-place experiments, but it omits the detailed reference-governor equations, objective functions, and low-level torque or position laws. A conventional flight-control design is said to be assumed but omitted. The result is a literature record that clearly specifies the modeling abstractions and infrastructure, while leaving the complete closed-form controller undocumented.

5. WAIR-inspired incline locomotion and coordinated thrust–contact control

The most technically developed extension of Husky Carbon is thruster-assisted incline walking, explicitly inspired by wing-assisted incline running (WAIR) in chukar birds. The relevant papers treat the key problem as one of maintaining feasible ground reaction forces on steep slopes while respecting friction limits; the distinctive Husky capability is that the robot can add body forces through four ducted fans rather than relying only on posture and leg placement (Salagame et al., 2023, Krishnamurthy et al., 2024).

The 2023 collocation paper introduces a high-fidelity MATLAB SimScape model with 18 degrees of freedom and 13 distributed mass elements, together with a reduced-order model, HROM, in which the legs are massless and the torso is a 6-DOF rigid body. The full thruster-equipped system is reported as 7.6 kg, with the propulsion unit contributing 3.3 kg, and the four fans are described as each generating about 2 kg of thrust, for a total of roughly 8 kg (Salagame et al., 2023). The associated thesis states the friction-feasibility condition in the standard form

$$\frac{F_h}{F_z} < \mu, \qquad F_z > 0, \qquad F_h = \sqrt{F_x^2 + F_y^2},$$

which captures the requirement that tangential force remain below frictional support while contact remains compressive (Wang, 2023).

Control synthesis is posed as a finite-horizon optimization in which joint accelerations, ground reaction forces, and thruster forces are stacked into a single control vector and optimized by a cubic Lobatto collocation method solved with MATLAB fmincon (Salagame et al., 2023). The 2024 paper uses a closely related collocation formulation and reports a 30-degree slope result with $\mu = 0.35$, emphasizing that without thrust the normal reaction can become close to zero or even negative, whereas optimized thrust keeps the active stance contact friction-admissible (Krishnamurthy et al., 2024).

The broader WAIR program, however, is reported more aggressively in the 2023 thesis and the 2023 collocation paper, which simulate inclinations of 0°, 10°, 20°, 30°, and 45°, with the headline result that Husky can walk up a 45-degree incline in simulation (Wang, 2023, Salagame et al., 2023). These papers also report an instructive force-sharing trend: torque peaks are higher at lower inclinations, because the legs carry more of the body weight there, whereas at higher inclinations the thruster contribution becomes larger, alleviating leg loading and keeping contact forces within friction-feasible regions.

6. Experimental status, limitations, and place within legged–aerial robotics

Experimentally, the most solidly documented Husky Carbon results remain terrestrial and partially assisted rather than fully multimodal. The 2022 progress report describes a test arena with multiple cameras, a textured padded surface, a Speedgoat target computer, an external power supply, and a large fan for cooling. The robot is tested in a tethered configuration with support ropes and a pulley mechanism that can alter effective leg loading. Within that setup, the team reports three-contact and two-contact trotting in place at different gait speeds, with phase portraits interpreted as convergence to a stable limit cycle (Salagame et al., 2022). The thesis later reports Propulsion Unit tests and thruster-assisted trotting in place, but not a completed hardware demonstration of sustained WAIR on physical inclines (Wang, 2023).

A major bottleneck across the Husky Carbon literature is compliance. The lightweight body and compliant legs induce body sagging at touchdown, reduce mechanical bandwidth, complicate impedance control and torque projection, and can generate large-amplitude oscillations. The narrow torso also reduces the support polygon. The progress report gives a particularly concrete consequence: when the gait cycle time exceeds approximately one third of a second, i.e. around a 3 Hz gait cycle, the projected COM can approach the support-polygon boundary even with three contacts, and stability degrades further without rapid corrective action. The authors explicitly state that they intend to address this by predicting the compliance and body center of mass using a model-based approach for more accurate kinematic and dynamic stability (Salagame et al., 2022).

Within the broader literature, Husky Carbon is positioned against LEONARDO, BALLU, and DUCK, which are bipedal thruster-assisted robots, and against wheeled or driving-flying hybrids such as DrivoCopter and FCSTAR. It is also contrasted with quadrupeds such as MIT Cheetah and ANYmal. The stated novelty is therefore not propeller augmentation per se, but a quadrupedal legged–aerial morphology that prioritizes low mass and aerial compatibility while accepting severe penalties in stiffness, support-polygon size, and terrestrial robustness (Salagame et al., 2022).

Two disambiguations are useful. First, Husky Carbon is unrelated to the Clearpath Husky A200 mobile base discussed in a separate configuration paper (Hiago et al., 2024). Second, it is unrelated to the theoretical carbon allotrope H-carbon from condensed-matter physics (He et al., 2012). In the robotics literature, “Husky Carbon” refers specifically to Northeastern University’s custom legged–aerial platform.

Taken together, the published record supports a precise characterization: Husky Carbon is a serious but unfinished hybrid robotics platform that has progressed from morphology-centered co-design, to integrated custom hardware and tethered trotting, to WAIR-inspired reduced-order control and steep-slope simulation. It has not yet been experimentally demonstrated as a fully autonomous flying quadruped with reliable mode transitions and onboard untethered operation.