Husky v.2: Legged–Aerial & UGV Robotics

• Husky v.2 comprises dual platforms: a quadrupedal robot with repurposable legs for both dynamic trotting and quadrotor flight, and a differential-drive UGV designed for heavy payloads and mobile manipulation.
• The legged–aerial system employs structure repurposing and integrated thrust–posture optimization using kinematic, dynamic, and NLP control frameworks to achieve narrow-path stability and agile mode switching.
• The UGV version offers a modular ROS2-based architecture with extensive sensor and manipulator integration, supporting long-duration, field-deployable operations in robotics research.

Husky v.2 refers to two distinct but widely referenced robotic research platforms: (1) Northeastern University’s multi-modal legged–aerial robot designed for dynamic quadrupedal locomotion and structure-repurposed flight, including its “Husky Carbon” narrow-path variant; and (2) Clearpath Robotics’ Husky A200 (v.2), a commercial, large-payload differential-drive UGV frequently integrated with manipulators for mobile manipulation. Both have become canonical platforms in multi-modal mobility and robotics research.

1. Structure-Repurposed Legged–Aerial Husky v.2: Definition and Key Objectives

Husky v.2, as described by the Northeastern University team, is a four-legged quadrupedal robot capable of both dynamic trotting and morphing into a hovering quadrotor by repurposing its leg architecture as flight arms. The platform deploys “structure repurposing”—using the same hip, femur, and tibia/fibula elements for both support and aerial propulsion—to address the critical trade-off between sufficient ground mobility and high thrust-to-weight ratio for lift. The system integrates posture manipulation with thrust vectoring, directly targeting the modal conflicts that typically prevent efficient ground-aerial transformations in robotics (Wang et al., 10 Oct 2025).

2. Mechanical and Electronic Architecture

Husky v.2: Northeastern Multimodal Leg–Rotor Robot

• Mechanical design: Each of the four legs consists of three actuated joints: hip frontal (roll), hip sagittal (pitch), and knee (tibia–fibula linkage). Legs are constructed from off-the-shelf carbon-fiber tubes with Onyx 3D-printed joints. During aerial flight, legs lock into horizontal orientations, forming quadrotor arms (no additional booms or arm extensions are present).
• Actuators & propulsion:
  • Hip roll/pitch: Dynamixel XH540-W270-T servos (9.9 Nm stall torque).
  • Knee: Dynamixel XM540-W270-T (10.6 Nm).
  • Propulsion: Four SunnySky X4112S brushless motors with EOLO 50A ESCs and 14×4.7″ propellers.
• Power and computation:
  • 6S LiPo battery (22.2 V, 240 g) provides ≈13.4 kgf peak thrust (thrust:weight ≈2).
  • Electronics include Cube Orange+ for flight control, NVIDIA Jetson Orin Nano for onboard vision/planning, and a U2D2 bus for actuator networking.
• A200 UGV Platform (Editor’s term): Key parameters for Clearpath Husky A200 (v.2) include 990×670×544 mm footprint, ≈50 kg curb weight, up to 75 kg payload, differential drive, fused 24 V power rails, modular ROS2 stack, and extensive sensor and manipulator integration via T-slot chassis (Hiago et al., 2024).

3. Locomotion, Posture Manipulation, and Thrust Integration

Ground–Aerial Mode Switching (Northeastern Husky v.2)

• Mode 1: Legged (Trotting) Locomotion
  • Two-contact diagonal-support trotting (Raibert-style control) with foot placement updated per CoM velocity error.
  • Thrusters provide minor roll stabilization by vectoring small thrusts during rapid maneuvers; reduces foot placement correction burden (Wang et al., 10 Oct 2025).
• Mode 2: Aerial (Quadrotor)
  • Morphing sequence: crouch, outward swing of frontal hips, realignment of props vertically above CoM, and takeoff within ≈10 s.
  • No additional flight structures: the legs themselves are flight arms, minimizing mass penalties and maximizing thrust-to-weight.

Posture–Thrust Synergy for Narrow-Path Gaits

• Husky v.2 can dynamically balance on narrow supports (pipes, slacklines) by leveraging both posture adjustments and body-stabilizing thrusts, as modeled in the Husky Reduced-Order Model (HROM) (Krishnamurthy et al., 2024).
• Thrust vectoring complements leg ground-reaction stabilization for tasks requiring fine lateral/roll control but limited support area, a feature validated in simulation for narrow-path gaits.

4. Kinematic, Dynamic, and Optimal Control Frameworks

Quadrupedal–Aerial Model (Northeastern)

• Leg kinematics: For each leg, forward kinematics $x_i = f(q_i)$ and inverse kinematics $q_i = f^{-1}(x_{i,\mathrm{des}})$, with Jacobian $J_i = \partial x_i/\partial q_i$ mapping desired end-effector forces to joint torques.
• Legged dynamics (standard floating-base): $$M(q)\,\ddot q + C(q,\dot q)\,\dot q + G(q) = S^T\,\tau + J^T\,F_\mathrm{grf}$$.
• Flight dynamics: Rigid body (SO(3)) with torque control $$I\,\dot\omega + \omega \times I\,\omega = M_\mathrm{ctrl}$$ and PD or cascade PID loops for attitude.
• Mixer for thrust: $u = [f_1, ..., f_4]^T$, distributed via a standard mixing matrix.

Integrated Thrust–Posture Optimization (Narrow Path)

• HROM framework treats torso as floating, with massless legs imparting GRFs and external body wrench $u_t$ from thrusters. Dynamics:

$$D(q)\,\dot v + C(q,v)\,v + G(q) = \sum_{i\in F} B_{g i} u_{g i} + u_t$$

• Ground interactions use a compliant contact plus Stribeck friction model; trajectories are parameterized by cubic polynomials per collocation interval.
• NLP formulated to steer body/leg states $x_k$ to references $x_k^{\rm ref}$ while keeping thruster wrenches and attitude errors penalized: $q_i = f^{-1}(x_{i,\mathrm{des}})$0.
• Extraction of individual thruster commands $q_i = f^{-1}(x_{i,\mathrm{des}})$1 by pseudo-inverting a wrench mapping $q_i = f^{-1}(x_{i,\mathrm{des}})$2 matrix derived from thruster locations and directions (Krishnamurthy et al., 2024).

5. Experimental Performance and Quantitative Results

Husky v.2 Multimodal (Northeastern)

• Trotting: Untethered duration of 8 s at ≈0.4 m/s; Raibert-style push recovery with ≈5 N·s impulse; rms CoM velocity tracking error ≈0.05 m/s.
• Morph/hovering: Leg-to-flight transition ≈10 s, with 20 s stable hover per 6S LiPo pack; roll/pitch/yaw hover std dev: 2°, 1.8°, 3°, respectively; thrust-to-weight ratio ≈2.
• Energy: Single battery charge yields ~20 s hover; payload experiments (e.g., 200 g) are planned but unreported.
• Narrow-path dynamic walking (in simulation): ≈0.1 m/s mean forward velocity over 3.5 s; CoM excursions confined within a few cm; angular rates <0.02 rad/s; EDF thrust saturations held ≲10% of the time (Wang et al., 10 Oct 2025, Krishnamurthy et al., 2024).

Clearpath Husky A200 v.2 (Differential-Drive UGV)

• Typical run time (3 h at ≈100 W avg. draw, 264 Wh battery) for sensor and compute-intensive missions; up to 75 kg payload and full ROS2/MoveIt stack for integrated perception and mobile manipulation (Hiago et al., 2024).

6. Control Software and Configuration

Stack Element	Husky v.2 Legged–Aerial	Husky A200 (Clearpath)
OS/firmware	Custom, Cube Orange+, Jetson, ROS	Ubuntu 22.04 + ROS2 Humble
Joint/network	Dynamixel SDK, U2D2 bus	ROS2 husky_base node, tf2
Perception	Not present (planned)	Velodyne LiDAR, RealSense
Planning/Control	Raibert (trotting), PID/SO(3) (flight), HROM NLP (sim)	Nav2, MoveIt, UR5 driver

Typical Husky A200 workflow includes sequential ROS2 launches for drive, perception, MoveIt-based manipulation, and UR5+gripper control, with safety checks and parameter tuning mandated (Hiago et al., 2024). Husky v.2’s ground–flight transitions remain operator-triggered; future automatic perception-enabled reconfiguration is identified as a research target (Wang et al., 10 Oct 2025).

7. Limitations, Comparative Insights, and Future Directions

Husky v.2 Multimodal Legged–Aerial

• Strengths: Structure repurposing minimizes flight mass, yielding high thrust-to-weight and credible dynamic trotting plus aerial hover in a unified architecture. Thruster-assisted stabilization adds new behaviors (e.g., pipe walking, simulated slacklining).
• Limitations: Current transitions are manual and slow (∼10 s); morphing is not automatic. Onyx 3D-printed legs exhibit compliance that limits high-speed ground performance. Lack of onboard perception or autonomy precludes deployment in unstructured environments.
• Future avenues: Onboard perception integration, unsupervised autonomous morphing, unified leg–thruster control (for assisted stair climbing, real dynamic slack-line traversal), and propeller-arm repositioning for redundancy and agile flight are specifically prioritized (Wang et al., 10 Oct 2025).

Narrow-Path Control Framework (Husky Carbon)

• Validated: Collocation-based optimal control is feasible for simulated narrow-path gaits; continuous-time feasibility and modest decision-variable counts are achieved.
• Unresolved: Lack of hardware validation and inability to simulate non-flat terrain or pipe geometries; computational overhead of trajectory optimization blocks real-time adaptivity (Krishnamurthy et al., 2024).

Husky A200 (UGV)

• Scope: Standard platform for ground-based mobile manipulation; fully documented sensor, actuator, and software configuration workflows assure reproducibility and ease of integration with manipulators like UR5 + Robotiq gripper (Hiago et al., 2024).
• Role in field: Does not target aerial/legged multimodal capability; platform is designed for terrestrial, payload-centric robotics applications.

Both lines of Husky v.2 research exemplify leading-edge platforms in their respective modalities—either as pioneers in hardware morphing integration for legged–aerial mobility or as robust, extensible bases for research in field robotics and mobile manipulation.