Multi-Modal Mobility Morphobot (M4) with appendage repurposing for locomotion plasticity enhancement

The study addresses how to achieve extensive locomotion plasticity in a single mobile robot capable of negotiating unstructured, multi-substrate environments (ground and air). The central hypothesis is that biomimetic appendage repurposing—using the same physical appendages as wheels, legs, hands, and thrusters—can overcome conflicting design requirements and enhance scalability and autonomy. The work introduces the Multi-Modal Mobility Morphobot (M4), motivated by search-and-rescue scenarios where environments may be partially collapsed, flooded, or obstructed. In such settings, aerial mobility offers rapid situational awareness, while diverse ground modes (wheeled, crouching, balancing/two-wheel mobile inverted pendulum, legged, tumbling) enable navigation through confined or rugged areas. The purpose and significance lie in demonstrating that manipulation of redundancy via morphing appendages can maximize locomotion modes, improve payload scalability, and enable autonomous multi-modal operations within one platform.

The paper frames prior work through three views. View 1 (Morpho-Functionality): multi-modal locomotion via morphing articulated structures (rigid or soft), enabling appendages to change function (e.g., legs, wings, wheels, flippers). While many designs (legged, slithering, amphibious, reconfigurable joints, morphing multirotors, shape-shifting wheels) exist, they typically realize only 2–3 modes and face scalability limits, especially in soft robots due to power and force density constraints. View 2 (Redundancy): multi-functionality achieved by adding separate appendages for single functions on non-morphing bodies (e.g., wheeled-aerial or legged-aerial hybrids). These are simpler but limited in number of modes and burdened by added mass. View 3 (Manipulation of Redundancy by Morphing): inspired by animals that repurpose appendages to create or eliminate redundancy as needed (e.g., sea lions flipper-assisted walking, meerkats bipedal scouting, Hoatzin nestlings wing-assisted quadrupedal locomotion, Chukar birds’ wing-assisted incline running). Some prior robotic demonstrations (flipper-leg or wheel-leg repurposing) show limited diversity and scalability. The authors position M4 as exploiting redundancy manipulation more extensively: the same four appendages can be repurposed into combinations of legs, wheels, hands, and thrusters, enabling eight modes and strategies like WAIR-inspired slope climbing and tumbling over large obstacles.

Design and system overview: M4 features an articulated body with four legs; each leg has two actuated hip joints (frontal and sagittal) and an integrated shrouded propeller that simultaneously functions as a wheel and a thruster. Frontal joints enable sideways leg rotation; sagittal joints enable forward/backward swing. The shroud acts as a wheel driven by a rim gear, while a propeller-motor inside the shroud provides thrust along the wheel axis. Considering propellers and shrouds, the robot has 16 actuators; total DOFs including body pose are 22. The robot weighs ~6.0 kg (5.6 kg in detailed table without stereo camera), measures 0.7 m (L) × 0.35 m (W/H) in UGV mode; 1.0 m tall in MIP mode; 0.3 m tall in UAS configuration with propeller centers up to 0.45 m apart. Each propulsor provides up to ~2.2 kgf, totaling ~9 kgf thrust. Wheels are 0.25 m diameter; leg (including wheel) length 0.3 m. Carbon fiber and Onyx-based fiber-reinforced 3D printed parts are used. Electronics include two microcontrollers (posture/wheel and thruster control), a high-level computer (e.g., Jetson Nano) for autonomy, sensors (encoders, IMU, stereo/depth cameras), communications, power electronics, and battery. Design rationale: Adopt appendage redundancy manipulation via repurposing to share mass across modes and enable force amplification by recruiting heterogeneous actuators (e.g., using thrust to augment traction on slopes) or homogeneously (repurposing all appendages to thrusters to increase thrust-to-weight). Control and modeling: A unified Euler–Lagrange model is derived for UGV, MIP, UAS, thruster-assisted MIP/WAIR, legged locomotion, and loco-manipulation. Ground interaction uses Stribeck friction and compliant contact models (spring-damper normal forces). The dynamics are expressed in state-space form ẋ = f(x) + g(x)u, where inputs include joint torques, wheel tractions, thruster forces, and ground contact resultant forces. The MIP/WAIR problem is solved with a direct collocation nonlinear dynamic programming approach. The cost function penalizes state tracking and control effort across discretized time; constraints include collocation consistency, boundary conditions, input limits, and friction cone constraints. Inputs are linearly interpolated per interval; states use cubic interpolation satisfying derivatives at interval boundaries and midpoints. This enables real-time feasible control for dynamic maneuvers like uprighting and slope ascent. Path planning (MM-PRM + A*): A multi-modal PRM is constructed with separate node sets for ground (z = z_g) and air (z ≥ z_g, z > 0), with parameters R = 4 m, N_g = 300, N_f = 300. Edges connect nearby nodes within radius R without obstacle intersections. Edge costs include ground locomotion energy C_r = ∫ P_w dτ, flight energy C_f = P_f + m g (z2 − z1), and transition energy C_t = ∫ P_t dτ. A* searches the graph with heuristic combining straight-line ground driving and vertical flight approximations, minimizing total energy while selecting modes and transitions. Experiments: Multiple experiments demonstrate modes and autonomy. Teleoperated UGV–UAS transition crossing a pond; closed-loop MIP uprighting and descent using collocation-based controller with desired angular rates of 10°/s (start) and 5°/s (near stand-up completion); crouching under low clearance; MIP-based grasping manipulation with free appendages; quadrupedal walking on rough terrain (wheels locked, alternating diagonal swing/stance; no knees). Autonomy tested first with motion capture for state estimation, then fully onboard (Jetson Nano + Intel RealSense) performing MM-PRM + A* to generate and follow waypoints, including takeoff/landing and ground navigation. WAIR-inspired thruster-assisted MIP ascents on 45° slopes; tumbling over large obstacles via sequential thruster-assisted uprighting and rolling motions.

• Modal diversity: M4 executes eight modes by appendage repurposing: fly (UAS), roll (UGV), walk (quadrupedal), crouch, balance (MIP), tumble, scout (elevated sensing in MIP), and loco-manipulate (using free appendages in MIP). - Steep slope traversal: Demonstrated WAIR-inspired thruster-assisted MIP climbing on a 45° slope, which is infeasible with pure UGV or legged modes. - Obstacle negotiation: Demonstrated tumbling/vaulting over large obstacles via thruster-assisted uprighting sequences. - Autonomous multi-modal operation: MM-PRM + A* path planning achieved multi-mode trajectories using motion capture, including landing on a 1.4 m tall platform and morphing back to UGV. Fully onboard autonomous multi-modal navigation with Jetson Nano and Intel RealSense generated waypoints, detected unreachable ground waypoints, autonomously morphed to UAS to overfly obstacles, landed, and resumed UGV to the goal. - Closed-loop MIP control: Collocation-based controller achieved stable UGV↔MIP transitions with desired angular rate profiles (10°/s at start of uprighting, 5°/s near completion), enabling dynamic balancing. - Hardware performance: Total weight ~6.0 kg; dimensions: UGV 0.7×0.35×0.35 m; MIP height 1.0 m; UAS height 0.3 m; propulsor spacing up to 0.45 m. Each propulsor provides up to ~2.2 kgf; total thrust ~9 kgf, supporting flight with payload margin. - Energy/power characteristics: Wheel motors 12 V at 1–3 A; propeller motors 24 V at 20–40 A; joint servos 7.4 V at 0.1–0.3 A. Estimated electrical power by mode/environment (representative): UGV flat ~30–100 W; legged ~40 W; MIP ~50 W (excluding transitions); crouching ~45–65 W; UAS ~3000 W; tumbling and thruster-assisted MIP consume less than continuous UAS (e.g., 1000–1500 W on ramps; MIP upward transition costs ~1500 W per transition). - Scalability: Repurposing shares mass across modes and allows homogeneous recruitment of appendages to scale thrust-to-weight (e.g., quadrupling when switching from UGV to full UAS), supporting payload capacity for onboard sensors and computing. - Quadrupedal mobility: Demonstrated walking on rough terrain by locking wheels and alternating swing/stance legs (limited without knees). - Manipulation: Demonstrated grasping with free appendages in MIP mode for loco-manipulation.

The findings validate that morphing-based manipulation of redundancy enables a single platform to overcome conflicting requirements of ground and aerial mobility. By repurposing appendages, M4 increases effective force capacity where needed: thrust augments traction for slope climbing (WAIR analogue) and provides height advantage for tumbling over large obstacles. Sharing components across modes improves scalability compared to redundancy-only or pure morpho-functional designs, letting M4 carry onboard computation and sensing for autonomous multi-modal operation. Energy analyses underscore the practical significance: UGV is far more efficient than UAS, so the planner chooses ground motion when feasible and reserves flight for impassable regions. Hybrid strategies (thruster-assisted MIP, tumbling) bridge capability gaps at lower energy cost than full flight. Autonomy results show M4 can perceive, plan, and execute multi-modal transitions end-to-end, addressing the research aim of versatile, self-contained operation in complex environments like search and rescue. Overall, M4 demonstrates unprecedented locomotion plasticity among multi-modal robots, substantiating the design hypothesis that appendage redundancy manipulation enables broad modal diversity and payload scalability.

This work introduces M4, a morphobot that repurposes its appendages as wheels, legs, hands, and thrusters to realize eight locomotion modes with autonomous, self-contained operation. Experiments demonstrate ground–air transitions, closed-loop MIP balancing, crouching, rough-terrain quadrupedal locomotion, thruster-assisted WAIR on 45° slopes, tumbling over large obstacles, and autonomous MM-PRM + A* multi-modal navigation with onboard sensing and computing. The approach leverages redundancy manipulation to share mass across modes and boost force capacity, yielding scalability and capability beyond prior designs. Future research directions include: adding leg DOFs (e.g., knees) for dynamic natural gaits; expanding manipulation beyond grasping (e.g., tool use); developing comprehensive decision-making to autonomously select and transition among all modes in more complex environments; enhancing onboard mapping/traversability analysis (e.g., robust occupancy mapping) and leveraging elevated MIP scouting for improved perception.

• Manipulation capability is limited to simple grasping with free appendages in MIP mode. - Leg architecture lacks knee joints, restricting natural quadrupedal gaits and dynamic legged performance. - Autonomous decision-making for mode selection is currently limited; full autonomy exists for UGV↔UAS transitions, while broader multi-mode switching requires further development. - UAS mode is energy-inefficient (~3 kW), constraining endurance; hybrid maneuvers mitigate but do not eliminate energy limitations. - Some demonstrations (e.g., pond crossing) were teleoperated; fully autonomous tests were conducted in controlled lab environments, potentially limiting generalizability to unstructured outdoor settings. - MIP transitions incur significant peak power (~1500 W per upward transition), which may limit frequent use under tight energy budgets.