BAXTER: Bi-modal Aerial-Terrestrial Hybrid Vehicle for Long-endurance Versatile Mobility: Preprint Version
Hyungho Chris Choi, Inhwan Wee, Micah Corah, Sahand Sabet, Taeyeon Kim, Thomas Touma, David Hyunchul Shim, Ali-akbar Agha-mohammadi
BBAXTER: Bi-modal Aerial-Terrestrial HybridVehicle for Long-endurance Versatile Mobility:Preprint Version
Hyungho Chris Choi , Inhwan Wee , Micah Corah , Sahand Sabet , TaeyeonKim , Thomas Touma , David Hyunchul Shim , and Ali-akbarAgha-mohammadi Korea Advanced Institute of Science and Technology, Republic of Korea Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA Univesity of Arizona, Tucson, AZ, USA
Abstract.
Unmanned aerial vehicles are rapidly evolving within thefield of robotics. However, their performance is often limited by pay-load capacity, operational time, and robustness to impact and collision.These limitations of aerial vehicles become more acute for missions inchallenging environments such as subterranean structures which may re-quire extended autonomous operation in confined spaces. While softwaresolutions for aerial robots are developing rapidly, improvements to hard-ware are critical to applying advanced planners and algorithms in largeand dangerous environments where the short range and high suscepti-bility to collisions of most modern aerial robots make applications inrealistic subterranean missions infeasible. To provide such hardware ca-pabilities, one needs to design and implement a hardware solution thattakes into the account the Size, Weight, and Power (SWaP) constraints.This work focuses on providing a robust and versatile hybrid platformthat improves payload capacity, operation time, endurance, and versa-tility. The Bi-modal Aerial and Terrestrial hybrid vehicle (BAXTER) isa solution that provides two modes of operation, aerial and terrestrial.BAXTER employs two novel hardware mechanisms: the
M-Suspension and the
Decoupled Transmission which together provide resilience duringlanding and crashes and efficient terrestrial operation. Extensive flighttests were conducted to characterize the vehicle’s capabilities, includingrobustness and endurance. Additionally, we propose Agile Mode Trans-fer (AMT), a transition from aerial to terrestrial operation that seeksto minimize impulses during impact to the ground which is a quick andsimple transition process that exploits BAXTER’s resilience to impact.
Problem Statement and Challenges:
Consider an autonomous robot travers-ing through a large and complicated subterranean environment with narrowcorridors, prone to impact, and disconnected stretches of ground-traversable ter-rain. In such an environment, both conventional drones and unmanned ground a r X i v : . [ c s . R O ] F e b Hyungho Chris Choi et al.: Preprint VersionFig. 1: Agile Mode Transfer. BAXTER exhibits a simple transfer from aerial to terres-trial operation via a short free fall with body inclined to reduce impulses to wheels. vehicles (UGV’s) would have limited scope of operation due to limited opera-tional time and lack of traversability, respectively.The problem mentioned aboveis a widespread challenge for traversing subterranean environments—includingmines, caves, and urban sprawl—which require extensive endurance to impactand versatile solutions that can traverse across rough and steep terrain [1]. Al-though potential functions of drones continue to expand by various algorith-mic advances [2,3], the development of hardware platforms that simultaneouslyachieve both the versatility of aerial vehicles and the endurance of ground vehi-cles is an important topic for hardware engineering in robotics. The challengesin developing such platforms include delicate management of payload and hard-ware features in limited size, weight, and power (SWaP) constraints, maintainingstructural and aerial stability, and all-around robustness of co-dependent com-ponents (e.g. coupled drive and transmission ) to mitigate risk of system failure.
Proposed Work and Contribution:
This publication outlines the devel-opment of the B i-modal A erial -TER restrial Hybrid Vehicle (BAXTER), shownin Fig. 1, which aims to provide a robust hardware-based solution for the afore-mentioned challenges. The highlights of the contributions are: – Improved Operating Scope and Range via the efficient use and transi-tion between two modes of operation: Terrestrial (low-versatility and high-efficiency) and
Aerial (high-versatility and low-efficiency). – Improved Resilience through introduction of wheel assemblies around theaerial propulsion units as well as the suspension system, emphasizing re-silience to impact while reasonably considering SWaP constraints. – Novel Mechanical Concepts (the
M-Suspension and
Decoupled transmis-sion ) were introduced to realize the contributions above. – Agile Mode Transfer demonstrates the versatility and resilience of theplatform through a simple transition from aerial to terrestrial operation.
Outline:
In this paper, novel mechanical concepts are proposed, realized,and tested to provide the above-mentioned contributions. Section 2 summarizesnotable developments in hybrid vehicles, morphing drone concepts, and general-purpose aerial vehicles. Section 3 states the goals and context of this paper.Section 4 explains the design concept and dynamics, and Sections 5 and 6 sum-
AXTER: Hybrid Aerial Vehicle: Preprint Version 3 (a) Rollocopter [4] (b) Drivocopter [6] (c) Fstar [7]
Fig. 2: Hybrid Vehicles and their Classifications. (a) Passive drive, non-morphing. (b)Active drive, non-morphing. (c) Active drive, active morphing. marize the realization and testing of the first prototype. Sections 7 and 8 provideinsight into the development process, propose future works, and conclude.
This section outlines previous hardware-based efforts on hybrid platforms, mor-phing drone concepts, and multi-functional aerial vehicles. Additionally, the endof this section highlights the design iterations toward developing the BAXTERrobot platform.
Hybrid Platforms:
Hybrid Platforms are a class of platforms that includemore than one mode of operation [2,4,5]. In the scope of this publication, un-manned hybrid platforms incorporating aerial and terrestrial operation are themain focus. The current Hybrid platforms can be categorized in two ways: – Separate Articulation in Terrestrial Operation : Platforms utilizingseparate means of propulsion in terrestrial mode are classified to have
ActiveTerrestrial Mode of operation. Vehicles with Active Terrestrial Mode [6,7],compared to vehicles with Passive Terrestrial Mode [2,4,5], provide a moreefficient and mutually independent operation, while their payload capacityis partially reduced due to the addition of the driving system. – Morphing Mechanism : Platforms utilizing separate built-in mechanismsfor changing the internal form factor are
Morphing platforms. Platformscan be
Active Morphing [7],
Passive Morphing , or
Non-Morphing [2,4,5].Morphing capabilities in hybrid vehicles provide vehicles with specializedforms for each operation, but may limit payload capacity or scale.
Morphing Drone Concepts
Apart from hybrid platforms, this class ofnovel aerial vehicle design focuses on morphing or controlled dynamic reconfig-uration of the aerial structure to increase efficiency or versatility. Single-entityapproaches manipulate the form of individual drones to achieve favorable char-acteristics. Examples include actuation of thrust vectors to achieve additionalfreedom in motion [8,9,10], translation of the thrusters to assume narrow profilesfor traversing tight spaces [11,12,13], along with other unique morphing concepts[14,15,13]. Multi-entity approaches arrange or integrate many drone units with
Hyungho Chris Choi et al.: Preprint Version (a) MODEL 1 “TUMBLER” (b) MODEL 2 “HEXWING” (c) MODEL 4 “BAXTER”
Fig. 3: Summary of Design Iterations. (a) The initial working prototype of protected hy-brid operation. (b) The initial working prototype for spherical wheel-rotor subassembly.(c) Incorporation of novel mechanical concepts to emphasize resilience and versatility. similar characteristics to achieve different forms and characteristics to producemulti-link chains [16,17] or other topologies.
All-around Aerial Vehicles
General purpose aerial vehicles represent an-other class of aerial vehicle designs that focus on SWaP (Size, weight, and power)constrained exploration challenges. Features of aerial vehicles specialized for ex-ploration in confined spaces include protective cages [18] and reconfiguration fortraversal of narrow passages [19]. Furthermore, the scope of aerial vehicle designsspans as far as extra-terrestrial environments [20,21]. Control systems for suchnewly developed vehicles are another topic of study to improve resilience [22] orefficiency [23].
The development of BAXTER followed the design, manufacture, and evaluationof two predecessor models: MODEL 1 “TUMBLER” and MODEL 2 “HEXWING” (illustrated in Fig. 3). Notable contributions of each design include: – MODEL 1 “TUMBLER” An initial working prototype for a robot with aspherical protective cage and active terrestrial operation. Limited in payloadcapacity and versatility (Fig. 3a). – MODEL 2 “HEXWING”
A proof of concept with spherical subassembliesaround each arm. Increased versatility in terrestrial operation. Limited inresilience, scalability, and payload capacity. (Fig. 3b) – MODEL 4 “BAXTER”
Incorporation of novel mechanical concepts (the
M-Suspension and the
Decoupled Transmission ) to emphasize resilience andversatility. Increased payload capacity. (Fig. 3c)
This section describes the objectives for the development of BAXTER withappropriate context. One iteration, MODEL 3, was designed and not produced.AXTER: Hybrid Aerial Vehicle: Preprint Version 5
Goal statement:
The development of BAXTER aims to provide a hardware-based solution for the following capabilities: (1) A hybrid vehicle that signifi-cantly increases operational scope and range by utilizing two modes of travel:
Aerial and
Terrestrial . (2) A collision resistant hardware platform to maintaineach mode and transition between them. (3) A robust, non-specific hardwareplatform for use in general, third party applications. (4) A controlled demon-stration of the transition from aerial to terrestrial mode. Due to energy expensesin flight, the main mode of operation for BAXTER is Terrestrial, and its aerialoperation is intended to be briefly used to traverse through environments that donot permit terrestrial operation. This addresses navigation through challengingsubterranean environments such as tunnels, subway stations, and caves, whichinclude tight and narrow spaces where many existing platforms would be proneto collision. Further, the design directives described above are in line with thechallenging environments from the DARPA Subterranean Challenge, which in-spired the development of BAXTER [1].
This section describes the BAXTER hardware in detail along with the simpledynamic basis for the demonstrative landing procedure.
Base Design : Toward realizing the capabilities discussed in Section 3, BAX-TER’s design falls into the class of robots with an active terrestrial mode and passive morphing , according to Section 2. The design is centered on a conceptof rotating, separately powered cages around each unit of propulsion (Fig. 4).For aerial operation, an X8 configuration is used with 7-inch rotors mountedon 2500KV class motors. For terrestrial operation, a single DC motor powerseach pair of wheels via the novel transmission system. The rendering and overalldimensions of BAXTER’s mechanical design are illustrated in Fig. 4 (a)-(b). Novel Concepts:
BAXTER features novel mechanical suspension and motor-power transmission systems. The need for reliable shock resistance for both dropimpact and terrestrial operation motivated the incorporation of a suspension sys-tem into the aerial platform. The
M-Suspension (Fig. 4-(c)) is a novel approachfor vertical shock reduction in a hybrid aerial vehicle. This new type of pas-sive suspension design focuses on ease of manufacturing based on 3D printingand minimizing the amount of additional weight. However, this suspension in-troduces a new obstacle: transmitting power through the moving joint duringactive terrestrial operation. The
Decoupled Transmission (Fig. 4-(d)) is a powertransmission that has a timing pulley on an axis coinciding with each suspensionjoint. This design keeps the total pitch line length of the timing belt constantat all angles of the suspension joint (so that the belt tension remains constantat all times). In conjunction with the suspension design, this completes the me-chanical concept to deliver motor power to each wheel through a moving, shockabsorbing system. The X8 configuration is a planar configuration of 8 rotors, arranged in 4 coaxialpairs. Hyungho Chris Choi et al.: Preprint VersionFig. 4: (a) Rendering of BAXTER’s Full Body Without Sensors/Processors (b) Overalldimensions of the platform (c) The M-Suspension design provides resilience duringlanding, crashes, and terrestrial operation (d) The Decoupled Transmission deliversmotor-power to each wheel through a moving shock absorbing system.
A simplified kinematic model of a quick and simple procedure for transfer fromaerial to terrestrial mode referred to as
Agile Mode Transfer (AMT) is discussedhere (see Fig. 1). AMT includes a controlled impact with the ground at an anglewhich provides a fast transition from aerial to terrestrial operation.The basis of the simplified kinematics of AMT is outlined below. Figure 5illustrates the four phases of the process (Free Fall, Initial Impact, Roll, andFinal Impact). The subscripts used in Fig. 5 indicate the number of the phaseassociated with given variable. In the following, (cid:126)v n and (cid:126)ω n respectively representthe translational velocity and the angular velocity of the center of mass (CoM)of the chassis at the end of the respective phase. It is assumed that the CoM ofthe system is located at the center of the chassis. The term (cid:126)h n represents thefinal height of the CoM, vector (cid:126)d connects the contact point of the Back Wheel to the CoM, and vector (cid:126)d connects the contact point of the Front Wheel to theCoM. The four phases shown in Fig. 5 are discussed below:1.
Free Fall : Vehicle keeps a positive landing angle ( θ ) and is subjected tofree-fall motion until making contact with the ground. Governing principle:parabolic motion. v x = v x v y = − (cid:112) g∆h (1)Here ∆h is the magnitude of the vertical displacement of the CoM duringfree fall, v x is the desired entrance velocity in x-direction, and g is thegravitational acceleration.2. Initial Impact : The first wheel (i.e.
Back Wheel ) makes contact with theground. Governing principles: Conservation of angular momentum and no-
AXTER: Hybrid Aerial Vehicle: Preprint Version 7Fig. 5: Simplified Kinematics Model for BAXTER’s Agile Mode Transfer (AMT). AMTmodels the landing procedure as four phases: 1. Free Fall, 2. Initial Impact, 3. Roll, 4.Final Impact. slip condition. (cid:126)d × (cid:126)v = (cid:126)d × (cid:126)v + I(cid:126)ω (cid:126)ω × (cid:126)d + (cid:126)ω × (cid:126)r = (cid:126)v (2)Here (cid:126)d is the displacement vector connecting the Back Wheel center to theCoM and (cid:126)r connects the contact point to the center of the
Back Wheel .3.
Roll : The first wheel rolls while the second wheel is in motion toward theground. Governing principles: conservation of energy and no-slip condition. mv + 12 Iw + mgh = 12 mv + 12 Iw (3) (cid:126)ω × (cid:126)d + (cid:126)ω × (cid:126)r = (cid:126)v (4)Here the terms (cid:126)ω , m , and I present the angular velocity, mass, and momentof inertia of the chassis, respectively.4. Final Impact : The second wheel (i.e.
Front Wheel ) makes contact with theground. Governing principles: Conservation of angular momentum, no-slipcondition for both wheels. (cid:126)d × (cid:126)v + I(cid:126)ω = (cid:126)d × (cid:126)v + I(cid:126)ω (cid:126)ω × (cid:126)r + (cid:126)ω × (cid:126)d = (cid:126)v (cid:126)v = (cid:126)v f (5)Given the desired entrance and exit velocity ( v , v f ), this model is used todetermine the optimal landing angle ( θ ) that minimizes the maximum impactwith the ground. To achieve this angle, the impact impulses at phase 2 ( (cid:126)I ) andphase 4 ( (cid:126)I ) are defined as: (cid:126)I = m(cid:126)v − m(cid:126)v (cid:126)I = m(cid:126)v − m(cid:126)v (6)Then, the maximum impact I max is defined as: I max = max( | (cid:126)I | , | (cid:126)I | ) (7)Finally, the landing angle ( θ ) that minimizes I max is calculated numerically. Hyungho Chris Choi et al.: Preprint Version
Preliminary flight tests with the prototype verified BAXTER’s viability. Now,in this section, additional tests (outlined below) verify the flight characteristics,versatility, and resilience of the platform.
Basic Flight and Endurance Tests:
Basic maneuvers (hover, rectangle,etc.) validate the flight characteristics of the prototype. The results from thesetests provide a quantitative evaluation (
Flight and Mission Time, MaximumSpeed, and Control Limits ) of the vehicle along with qualitative assessments.
Drop Test:
BAXTER’s suspension system is unique since it is designed toabsorb step-like impulses from impact with the ground on landing or crashing.So, to measure the suspension endurance, a drop test was performed with theprototype in the all-up configuration (4.2 kg total) and in an upright positionfrom incrementally increasing heights to obtain the following metrics (Drop Lim-its): (1) Maximum Intact Height , the maximum height where the prototype takesno irreversible damage, and (2)
Maximum Flyable Height , the maximum heightwhere the prototype can still maintain aerial operation.
Agile Mode Transfer Test:
The agile mode transfer experiments exploitand demonstrate the effectiveness of the hardware design. From the kinematicsmodel described in Section 4, a mobile landing procedure will be devised byproviding the parameters in Fig. 6, and calculating the optimum landing angle( θ ) that minimizes the load at impact in landing ( I max ). The implementationof this maneuver is accomplished via ROS and the attitude and velocity controlcapability of the flight controller [24], with input from onboard sensors (Fig. 7). Autonomous Flight System:
To operate a mission autonomously in ter-restrial and aerial mode, BAXTER uses odometry for velocity and attitude. At-titude estimation is done through the 9-axis IMU (Inertial Measurement Unit)within the flight controller. For velocity and position estimation, an Intel Re-alSense T265 tracking camera was linked to a PX4 via ROS (Robot Operat-ing System) running in the companion computer through VIO (Vision InertialOdometry). The mission controller for AMT runs on the companion computer(Nvidia Jetson TX2), which implements the 3 mission phases outlined in Fig. 6.
Summary of Prototype:
Finally, the first prototype of BAXTER presentsa tangible proof of concept for the mechanical design described in Section 4. Thedetails of the circuit-level hardware are listed in Fig. 7-(a).
This section provides the qualitative and quantitative outcomes of the designand testing of BAXTER, described in Section 5.
Basic Flight Parameters:
Table 1 presents the results of the
Basic Flightand Endurance Tests and
Drop Test outlined in Section 5. These results were All-up refers to the maximum payload configuration which uses additional weightsas a proxy for sensing and computation payloads.AXTER: Hybrid Aerial Vehicle: Preprint Version 9Fig. 6: Agile Mode Transfer Mission Diagram, Mission Controller Stages (which aredistinct from the phases of the landing process in Section 4) and Parameters. This figureand the attached target table (controlled set-points for each phase are highlighted inred) illustrates and describes targets for the AMP in the mission perspective with 3Stages: (1): Before Deployment, (2): Deployment (vehicle adjusts pose in free fall), and(3): Disarm (flight controller switches off)Table 1: Obtained Flight Parameters for BAXTER. (a) Obtained results from mea-surements (b) Obtained results from experiments (a)
Category Sub-Category MetricMass
Chassis 1.2 kgPropulsion System 1.2 kgMaximum Payload 1.8 kg
Inertia I xx (roll axis) 0.351 kg · m (All-Up) I zz (pitch axis) 0.125 kg · m I yy (yaw axis) 0.254 kg · m (b) Category Sub-Category MetricTime
Flight (Manual) 5 min
Limit
Flight (Autonomous) 4 minDrive (Drive Only) 30 min
Control
Aerial Speed 2 m/s
Limit
Terrestrial Speed 1 m/sRoll Angle 30 degPitch Angle 25 deg
Drop
Max. Intact 1.3 m
Limit
Max. Flyable 1.4 m obtained and verified over 3 trials. Inertial measurements were conducted withthe bifilar pendulum method [25].
Agile Mode Transfer Test:
With the verified maximum intact drop heightof around 1.3 m and control limits drawn from the previous experimental re-sults, agile mode transfer solutions (varying Entrance and Exit Velocity) thatwere within the control limits were tested and verified to be repeatable. Figure 8depicts the case of: v i = 1 . , v f = 0 . , h = 0 . with respect to potential landingangles ( θ ). It can be deduced that the landing angle of around 20 deg producesthe least maximum impulse on impact. Such optimal landing angles are calcu-lated internally to the mission controller and are implemented as a sequence ofcontroller stages described in Fig. 6. A representative frame-on-frame picture ofthis landing procedure is shown in Fig. 1. This section lists notable insights obtained regarding the overall flight charac-teristics and test results of BAXTER.
Overall Insight:
Much insight into suspension systems for hybrid platformswas gained through development and experimentation on BAXTER. To the bestof our knowledge, BAXTER is the first platform to incorporate a passive suspen-sion system into a hybrid aerial vehicle. Notably, BAXTER’s flight characteris-tics are affected by deflection of the suspension joints, but fine adjustment of theflight parameters allowed for reasonable performance in autonomous control.
Vibration Due to the Suspension System:
This was evident in low-frequency vibration of the vehicle (magnitude of Hz ) during preliminary flighttests. Although these vibrations can be mitigated by careful tuning of the flightcontroller, doing so restricts the control limits. Motor Translation due to Body Distortion:
As a side effect of deflectingpassive suspension joints, passive morphing occurs, and the thrusters translatealong with the wheel assemblies in flight. This introduces irregularities in flightbehavior due to distortions of the body that are not present in conventionalaerial vehicles. A variant of this phenomenon has been studied with respect to afolding drone where a motor-actuated morphing mechanism involving translatingthrusters introduced undesired displacement in flight behaviour [11]. A potentialresearch topic can be to dynamically change parameters in the control system tomitigate the deflection of the suspension system in order to expand the controllimits for BAXTER.
Insight on Resilience:
As intended, the results from the drop tests (seeSection 5) demonstrate the resilience of BAXTER in contact-laden environments
AXTER: Hybrid Aerial Vehicle: Preprint Version 11
Back WheelFront WheelMinimum Impulse
Fig. 8: Plot of expected front ( I ) and back ( I ) wheel impulse values, varying thelanding angle ( v i = 1 . m/s, v f = 0 . m/s, h = 0 . m ) using the kinematics modeldescribed in Section 4.1. The optimum landing angle, yielding the lowest maximumimpulse value, is highlighted with a yellow circle. and crash-prone operation. A notable takeaway from the drop tests is that formany times that the prototype took damage its onboard electrical componentsremained intact while primarily easily-replaceable 3D-printed parts took con-centrated damage. A potential future design objective cloud be to concentratestructural weak spots within the platform to easily accessible parts (Commer-cially available parts, 3D printed parts, etc.). Insight on Agile Mode Transfer:
Agile Mode Transfer tests were a suc-cessful demonstration of BAXTER’s potential for improved operational scopeand resilience. Overall, BAXTER was able to reliably perform Agile Mode Trans-fer (AMT) without irreversible damage to the hardware.
Summary of BAXTER:
BAXTER presents a novel hardware-based approachto aerial-terrestrial hybrid platforms. It also offers a solution for multiple fronts inchallenging environments where both resilience and extensive operational scopeand range are required. This publication has detailed the design philosophy andprototype of the BAXTER concept and successfully showed viability of its hybridcapabilities by evaluation of conventional flight characteristics and testing thenovel Agile Mode Transfer.
Future Works:
BAXTER provides a reliable platform to operate in largeand contact-prone environments where operation of conventional aerial robotswould prove to be infeasible. BAXTER also opens avenues for applications (suchas subterranean exploration) that exploit its resilience which is demonstratedwith the Agile Mode Transfer. Introducing a contact-flexible platform also openspotential for new software features that exploit contacts such as for computationof odometry [3] as well as risk-aware mapping [26] and planning [27],[28] .Further iteration on this design will continue to push the boundaries of hy-brid robot platforms. For BAXTER, the introduction of the
M-Suspension andthe
Decoupled Transmission provided notable improvements. But, as describedin Section 7, challenges in aerial mobility arise as a trade-off between including a passive suspension system for resilience and adverse changes to the vehicle’sflight characteristics. One hardware-based solution to this challenge is the intro-duction of a simple active suspension, specifically a suspension system that canbe switched on or off. This change to the suspension system will be a main focusfor the design of the next iteration (MODEL 5).
Acknowledgements
This work was supported by the Institute for Information & communicationsTechnology Promotion (IITP), funded by the Korean government (MSIP) (De-velopment of AI-powered Autonomous Drone for Complex Indoor Environment)under the guidance of Unmanned System Research Group of Korea AdvancedInstitute of Science and Technology (KAIST USRG). Further guidance wasprovided from the Jet Propulsion Laboratory, California Institute of Technol-ogy, under a contract with the National Aeronautics and Space Administration(80NM0018D0004). We want to thank Hyunjee Ryu, Brian Kim, and HanseobLee of KAIST USRG, along with Brett Lopez and Team CoSTAR of JPL forvarious technical support and fruitful discussions.
References
1. DARPA: Unearthing the subterranean environment (2020) (accessed June 3, 2020).2. Fan, D.D., Thakker, R., Bartlett, T., Miled, M.B., Kim, L., Theodorou, E., Agha-mohammadi, A.: Autonomous hybrid ground/aerial mobility in unknown environ-ments. In: Proc. of the IEEE/RSJ Intl. Conf. on Intell. Robots and Syst., Macau,China (November 2019)3. Lew, T., Fan, D., Tara, B., Santamaria-Navarro, A., Thakker, R., Agha-mohammadi, A.: Contact inertial odometry: Collisions are your friends. In: Proc.of the Intl. Sym. of Robot. Research, Hanoi, Vienam (October 2019)4. Sabet, S., Agha-mohammadi, A.a., Tagliabue, A., Elliott, D.S., Nikravesh, P.E.:Rollocopter: An energy-aware hybrid aerial-ground mobility for extreme terrains.In: IEEE Aerospace Conf., Big Sky, Montana (March 2019)5. Kalantari, A., Spenko, M.: Design and experimental validation of HyTAQ, a HybridTerrestrial and Aerial Quadrotor. In: IEEE Intl. Conf. on Robot. and Autom.,Karlsruhe, Germany (May 2013)6. Kalantari, A., Touma, T., Kim, L., Jitosho, R., Strickland, K., Lopez, B.T., Agha-mohammadi, A.: Drivocopter: A concept hybrid aerial/ground vehicle for long-endurance mobility. In: IEEE Aerospace Conf., Big Sky, Montana (March 2020)7. Meiri, N., Zarrouk, D.: Flying STAR, a hybrid crawling and flying sprawl tunedrobot. In: IEEE Intl. Conf. on Robot. and Autom., Montreal, Canada (2019)8. Zheng, P., Tan, X., Kocer, B.B., Yang, E., Kovac, M.: TiltDrone: A fully-actuatedtilting quadrotor platform. IEEE Robot. Autom. Letters (4) (2020) 6845–68529. Kamel, M., Verling, S., Elkhatib, O., Sprecher, C., Wulkop, P., Taylor, Z., Sieg-wart, R., Gilitschenski, I.: The Voliro omniorientational hexacopter: An agile andmaneuverable tiltable-rotor aerial vehicle. IEEE Robot. Autom. Mag. (4) (2018)34–4410. Ryll, M., Bülthoff, H.H., Giordano, P.R.: A novel overactuated quadrotor un-manned aerial vehicle: Modeling, control, and experimental validation. IEEETrans. Control Syst. Technol. (2) (2015) 540–556AXTER: Hybrid Aerial Vehicle: Preprint Version 1311. Falanga, D., Kleber, K., Mintchev, S., Floreano, D., Scaramuzza, D.: The foldabledrone: A morphing quadrotor that can squeeze and fly. IEEE Robot. Autom.Letters (2) (2018) 209–21612. Riviere, V., Manecy, A., Viollet, S.: Agile robotic fliers: A morphing-based ap-proach. Soft Robotics (5) (2018) 541–55313. Bucki, N., Mueller, M.W.: Design and control of a passively morphing quadcopter.In: IEEE Intl. Conf. on Robot. and Autom., Montreal, Canada (May 2019)14. Hedayati, H., Suzuki, R., Leithinger, D., Szafir, D.: PufferBot: Actuated expand-able structures for aerial robots. ArXiv (2008.07615) (2020)15. Zhao, N., Luo, Y., Deng, H., Shen, Y.: The deformable quad-rotor: Design, kine-matics and dynamics characterization, and flight performance validation. In: Proc.of the IEEE/RSJ Intl. Conf. on Intell. Robots and Syst., Vancouver, Canada(September 2017)16. Nguyen, H., Dang, T., Alexis, K.: The reconfigurable aerial robotic chain: Modelingand control. In: IEEE Intl. Conf. on Robot. and Autom., Paris, France (May 2020)17. Zhao, M., Kawasaki, K., Chen, X., Noda, S., Okada, K., Inaba, M.: Whole-bodyaerial manipulation by transformable multirotor with two-dimensional multilinks.In: IEEE Intl. Conf. on Robot. and Autom., Singapore (May 2017)18. Flyability: Elios 2: Intuitive indoor inspection (accessed January 6, 2021).19. Brown, L., Clarke, R., Akbari, A., Bhandari, U., Bernardini, S., Chhabra, P., Mar-janovic, O., Richardson, T., Watson, S.: The design of Prometheus: A reconfig-urable UAV for subterranean mine inspection. Robotics (4) (2020) 9520. Balaram, B., Canham, T., Duncan, C., Grip, H., Johnson, W., Maki, J., Quon, A.,Stern, R., Zhu, D.: Mars helicopter technology demonstrator. In: AIAA Atmos.Flight Mech. Conf., Kissimmee, Florida (January 2018)21. Bayard, D.S., Conway, D.T., Brockers, R., Delaune, J.H., Matthies, L.H., Grip,H.F., Merewether, G.B., Brown, T.L., Martin, A.M.S.: Vision-based navigationfor the NASA Mars helicopter. In: AIAA Scitech 2019 Forum, San Diego, Canada(January 2019)22. Liu, Z., Karydis, K.: Toward impact-resilient quadrotor design, collision character-ization and recovery control to sustain flight after collisions. ArXiv (2011.02061)(2020)23. Sabet, S., Poursina, M., Nikravesh, P.E., Reverdy, P., Agha-mohammadi, A.: Dy-namic modeling, energy analysis, and path planning of spherical robots on uneventerrains. IEEE Robot. Autom. Letters (4) (2020) 6049–605624. Meier, L.: MAVLINK common message set for offboard control (2020) (accessedJune 23, 2020).25. Then, J.W., Chiang, K.: Experimental determination of moments of inertia by thebifilar pendulum method. American Journal of Physics (1970) 53726. Agha-mohammadi, A., Heiden, E., Hausman, K., Sukhatme, G.: Confidence-richgrid mapping. The International Journal of Robotics Research (12-13) (2019)1352–137427. Agha-mohammadi, A., Agarwal, S., Kim, S.K., Chakravorty, S., Amato, N.M.:SLAP: Simultaneous localization and planning under uncertainty via dynamic re-planning in belief space. IEEE Trans. Robotics34