Principles of Multi-Modal Robotic Locomotion

155 concepts86 papers2 proposals

In mobile robotics, engineers have long faced a frustrating compromise known as the mobility-efficiency trade-off. Wheeled vehicles are masterfully efficient on flat ground, achieving an incredibly low Cost of Transport (CoT), but they are easily stopped by a single high step or gap. Conversely, legged systems can step over obstacles, and aerial drones can fly over them entirely—but legs consume significant energy just to stand or trot, and drones must expend massive power continuously simply to fight gravity.

To break this trade-off, a powerful paradigm has emerged: multi-modal locomotion. By physically and algorithmically blending distinct modes of transport—such as rolling, stepping, and flying—robots can dynamically choose the most efficient mechanism for the terrain at hand.

But achieving this is not as simple as bolting wheels onto a drone or a quadruped. Doing so introduces severe weight penalties and highly complex, non-linear control challenges. Understanding why modern multi-modal robots work requires looking at how researchers have unified hardware design and optimal control to make these transitions seamless.


Wheeled-Legged Fusion: Wheels as Moving Footholds

The most direct way to combine speed and traversability is to place wheels at the extremities of a legged robot. Early attempts treated these modalities as separate systems: the robot would either lock its legs and drive like a car, or stop its wheels and walk.

Modern architectures achieve a massive leap in efficiency by treating wheels as torque-controlled moving footholds within a unified mathematical framework. As demonstrated by C. Dario Bellicoso et al. in Keep Rollin' - Whole-Body Motion Control and Planning for Wheeled Quadrupedal Robots, integrating wheels directly into a whole-body dynamic optimization allows the robot to blend driving and stepping simultaneously. This approach yields an 80% to 85% reduction in energy consumption compared to traditional trotting gaits.

To understand why this works, consider how a standard quadruped moves. Every step requires accelerating and decelerating the mass of the leg, losing energy to impacts and joint braking. By utilizing a whole-body Model Predictive Control (MPC) framework, such as the one proposed by Ruben Grandia et al. in Whole-Body MPC and Online Gait Sequence Generation for Wheeled-Legged Robots, the system can keep its wheels in continuous contact with the ground on smooth patches, effectively turning the legs into an active suspension system. The optimization automatically discovers aperiodic gaits—stepping only when an obstacle violates a kinematic reachability limit, and driving whenever possible to minimize power consumption.

[Pure Driving: CoT ~ 0.1] ---> [Hybrid Stepping/Driving] ---> [Pure Walking: CoT ~ 1.5]
   (Flat, smooth terrain)         (Minor obstacles/slopes)        (Highly broken terrain)

Furthermore, this hybrid configuration unlocks unique physical synergies. For instance, Xuanqi Zeng et al. demonstrated in Jumping Control for a Quadrupedal Wheeled-Legged Robot via NMPC and DE Optimization that a robot can use its wheels to rapidly accumulate forward kinetic energy prior to a leap. When the robot hits a launch ramp or executes a jump, this pre-accumulated momentum allows it to clear significantly larger horizontal gaps than a purely legged robot jumping from a standstill.


Ground-Aerial Co-Design: Eliminating the Weight Penalty

Combining ground and aerial locomotion introduces a harsh physical constraint: the weight penalty. As noted by Junxiao Lin et al. in Skater: A Novel Bi-modal Bi-copter Robot for Adaptive Locomotion in Air and Diverse Terrain, adding heavy, dedicated drivetrains (like electric motors and gearboxes) for ground wheels severely degrades flight endurance, creating a self-defeating design loop.

To circumvent this, researchers utilize shared actuation and passive mechanics. The core insight is to let the primary aerial actuators (the rotors) do double duty as the ground propulsion system.

In Autonomous Hybrid Ground/Aerial Mobility in Unknown Environments, Tara Bartlett et al. present a "Rollocopter"—a quadrotor enclosed within or attached to passive wheel structures. Instead of driving the wheels with heavy internal motors, the robot tilts its rotors to generate forward aerodynamic thrust while remaining on the ground. Because the ground supports the robot's weight, the rotors only need to overcome rolling resistance rather than generating full vertical lift. This allows the system to consume up to five times less energy than flight alone.

Taking this concept further, Zhi Zheng et al. designed a system detailed in Roller-Quadrotor: A Novel Hybrid Terrestrial/Aerial Quadrotor with Unicycle-Driven and Rotor-Assisted Turning that embeds a quadrotor frame inside a single, large passive wheel. By utilizing differential rotor thrust to steer and roll the wheel, the vehicle achieves an operational duration over 40 times longer than continuous flight, while maintaining a slim profile capable of squeezing through narrow gaps.


Unified Planning Across Discrete Modes

The intelligence behind a multi-modal robot lies in its ability to decide when to switch modes. This is fundamentally a hybrid optimization problem: the robot must plan continuous trajectories (forces, velocities) while making discrete choices (e.g., "should I drive around this rock, step over it, or fly over it?").

Planning across these boundaries is computationally expensive. To achieve real-time performance, H.J. Terry Suh et al. in Energy-Efficient Motion Planning for Multi-Modal Hybrid Locomotion introduced methods that approximate the complex costs of switching between modes. By smoothing these discrete transitions into continuous optimization costs, robots can rapidly compute dynamically feasible paths through cluttered environments.

On the local level, this manifests as "anytime" planning. As shown by Tobias Klamt et al. in Anytime Hybrid Driving-Stepping Locomotion Planning, a robot can continuously evaluate its terrain perception. If the surface is flat, the planner commands omnidirectional driving. The moment an obstacle exceeds the radius of the wheel, the controller seamlessly transitions into a stepping gait, adjusting foot height based on local 3D maps without stopping the robot's forward progress.

                  [Obstacle Detected]
                           |
            Is obstacle height > wheel radius?
                     /           \
                  (No)          (Yes)
                   /               \
    [Maintain Rolling Mode]    [Switch to Stepping/Flight]
    Low CoT, High Speed        High CoT, Clears Obstacle

This multi-modal intelligence also requires specialized state estimation. When a wheel transitions from rolling to stepping, standard wheel odometry equations instantly fail due to slip and impact forces. As a result, frameworks like the one developed by Mojtaba Hosseini et al. in State Estimation for Hybrid Locomotion of Driving-Stepping Quadrupeds utilize "contact trust" models within Kalman Filters. These models dynamically down-weight wheel kinematics and rely more heavily on inertial sensors during transitions, ensuring the robot never loses track of its position during aggressive mode switches.

Alternatively, when a single platform cannot physically support multiple modalities, researchers deploy marsupial systems. Paolo De Petris in Marsupial Walking-and-Flying Robotic Deployment for Collaborative Exploration of Unknown Environments demonstrates this concept by using a high-endurance quadruped to carry a short-endurance, collision-tolerant drone. The ground robot navigates miles of rough terrain efficiently, and deploys the aerial scout only when encountering vertical chasms or multi-level structures, combining the best of both worlds through collaborative autonomy.

Key Foundational Work

Design of Disaster Relief Robot Assistant (2013) — contributed 3 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_hybrid_locomotion_8b08b427,
  title = {hybrid locomotion},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/hybrid-locomotion/rdr_8b08b427c5b04da5bd85c9aee7eda787},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_hybrid_locomotion_8b08b427_v1,
  title = {hybrid locomotion},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/hybrid-locomotion/rdr_8b08b427c5b04da5bd85c9aee7eda787/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake