ATMO: An Aerially Transforming Morphobot for Dynamic Ground-Aerial Transition

Abstract

Designing ground-aerial robots is challenging due to the increased actuation requirements which can lead to added weight and reduced locomotion efficiency. Morphobots mitigate this by combining actuators into multi-functional groups and leveraging ground transformation to achieve different locomotion modes. However, transforming on the ground requires dealing with the complexity of ground-vehicle interactions during morphing, limiting applicability on rough terrain. Mid-air transformation offers a solution to this issue but demands operating near or beyond actuator limits while managing complex aerodynamic forces. We address this problem by introducing the Aerially Transforming Morphobot (ATMO), a robot which transforms near the ground achieving smooth transition between aerial and ground modes. To achieve this, we leverage the near ground aerodynamics, uncovered by experimental load cell testing, and stabilize the system using a model-predictive controller that adapts to ground proximity and body shape. The system is validated through numerous experimental demonstrations. We find that ATMO can land smoothly at body postures past its actuator saturation limits by virtue of the uncovered ground-effect.

Figures (6)

• Figure 1: Proposed Dynamic Wheel Landing Maneuver Versus Quadrotor Landing. (A) Phases of the envisioned dynamic wheel landing maneuver classified according to altitude $z(t)$. The robot is initially in regular flight mode at some height $z_0$, as it descends it begins tilting its wheel-thrusters at some height $z_\varphi$. Finally, at some height $z^\ast$ the robot switches to the near ground controller and lands while maximizing the tilt angle $\varphi$. The flow field has been sketched to indicate the complexity of the aerodynamics at each phase. A free body diagram of the possible forces during morphing flight when the robot tries to compensate for roll disturbances is depicted in the top right corner. (B) Quadrotor transitioning to the ground and entering ground effect. The flow pattern is sketched qualitatively.
• Figure 2: Aerially Transforming Morphobot (ATMO) Overview (A) Illustration of ATMO in perspective with key components labeled. (B) Tilt mechanism. (left) ATMO is shown in a morphed configuration in flight. Underneath, the tilt actuator box is enlarged and the joints are labeled. The tilt mechanism is actuated by two co-rotating bevel gears actuated by a DC motor. The spinning causes joint $A$ to translate on the $OA$ axis, inducing mechanism motion. (right) The right half of the symmetric mechanism is shown with all the joints labeled as well as point $E$ which corresponds to the intersection of the propeller axis with the extension of $CD$. The path taken by joint $E$ as joint $A$ moves from bottom to top is traced in blue. Open colored joints represent rotating joints or sliders. $CDE$ is a unified link pivoted on D. The key kinematics parameter is the tilt angle $\varphi$ which varies from $\varphi=0$ in flight configuration to $\varphi=\frac{\pi}{2}$ in drive configuration. (C) Electronics architecture and connections. On the left are ground control components. These communicate to the onboard components by Wi-Fi or wireless (radio protocol) transmission. The onboard computer receives radio signals from the flight controller and communicates to all the motor drivers using ROS2.
• Figure 3: Morpho-Transition Aerodynamics.(A) Experimental setup. A 6 axis robotic arm is used to adjust the three dimensional position of ATMO. A load cell is in series between the robotic arm and ATMO. A laser sheet is positioned underneath the plane of the front two thrusters for imaging purposes. To achieve this a laser source is shone on a 45 degree mirror generating a vertical sheet which constitutes the imaging plane. (B) Smoke visualizations of the aerodynamic flow field with the robot stationary at tilt angles. For $\varphi=0^\circ$ two streams of air are flowing vertically through the wheel-thrusters. For $\varphi=30^\circ$ the two jets are reoriented and mix to form one downward oriented stream. At $\varphi=70^\circ$ the two jets impinge to form a stagnation point with an unstable region where the flow may be directed either upwards or downwards. Arrows indicated the overall direction of the flow as observed in the videos of the visualization (see supplementary materials). (C) Results from the load cell testing for $\varphi=40^\circ,50^\circ,60^\circ,70^\circ$. The overall thrust (measured at the load cell) normalized by the thrust produced in the same configuration far from the ground is plotted at 6 different heights $z=0.25,0.32,0.42,0.52,0.62,0.72 \textrm{m}$. The standard deviation of the thrust during the measured time horizon is shaded in lighter color for each experiment.
• Figure 4: Control Architecture. The control architecture is based on a central module which consists of two controllers: The flight-transition controller, and the drive controller. The control module receives reference points $x_{\textrm{ref}}, u_{\textrm{ref}}$ from a high level trajectory generator, and a position and orientation $\bm p, \bm \theta$ from a motion capture system/VIO system which is fed into the flight controller Extended Kalman Filter (EKF) which provides the state estimate $\hat{x}$. Using this information the flight-transition controller produces thruster actions $u_a$ and the drive controller produces wheel spin actions $u_g$. Whether the commands are actually sent to the thruster motor electronic speed controllers or wheel motor drivers depends on whether or not a ground contact/transition state have been estimated. The trajectory generator is responsible for producing the tilt velocity $v$ which is fed to the tilt actuator mechanism which in turn produces angle feedback, $\varphi$, to the controller using the mechanism kinematics and encoder count.
• Figure 5: Experimental Validation of Dynamic Wheel Landing. (A) Recorded data during an autonomous wheel landing. In blue $z(t),\varphi(t),\alpha(t),\bar{u}(t)$, i.e. the altitude, tilt angle, control blending factor, and the mean normalized thrust applied by the four rotors are plotted. The three vertical lines denote the time instances at which the robot begins to morph (height $z_\varphi$), the point at which transition begins (height $z^\ast$) and finally the point at which impact occurs (height $z_g$). The morphing, transition, and ground regions are highlighted in blue red, and green respectively. The morphing height, transition height, and ground contact height ($z_\varphi,z^\ast,z_g$) are indicated along with corresponding angles, and the maximum normalized thrust (B). Snapshots from the performed trajectory.