Hubbry Logo
Legged robotLegged robotMain
Open search
Legged robot
Community hub
Legged robot
logo
7 pages, 0 posts
0 subscribers
Be the first to start a discussion here.
Be the first to start a discussion here.
Legged robot
Legged robot
Legged robot
View on Wikipedia
from Wikipedia

Legged robots are a type of mobile robot which use articulated limbs, such as leg mechanisms, to provide locomotion. They are more versatile than wheeled robots and can traverse many different terrains, though these advantages require increased complexity and power consumption. Legged robots often imitate legged animals, such as humans or insects, in an example of biomimicry.[1] [2]

Gait and support pattern

[edit]

Legged robots, or walking machines, are designed for locomotion on rough terrain and require control of leg actuators to maintain balance, sensors to determine foot placement and planning algorithms to determine the direction and speed of movement.[3][4] The periodic contact of the legs of the robot with the ground is called the gait of the walker.

In order to maintain locomotion the center of gravity of the walker must be supported either statically or dynamically. Static support is provided by ensuring the center of gravity is within the support pattern formed by legs in contact with the ground. Dynamic support is provided by keeping the trajectory of the center of gravity located so that it can be repositioned by forces from one or more of its legs.[5]

Types

[edit]

Legged robots can be categorized by the number of limbs they use, which determines gaits available. Many-legged robots tend to be more stable, while fewer legs lends itself to greater maneuverability.

One-legged

[edit]

One-legged, or pogo stick robots use a hopping motion for navigation. In the 1980s, Carnegie Mellon University developed a one-legged robot to study balance.[6] Berkeley's SALTO is another example.[7][8][9][10]

Two-legged

[edit]
Main articles: Humanoid robot and Chicken walker
ASIMO - a bipedal robot

Bipedal or two-legged robots exhibit bipedal motion. As such, they face two primary problems:

  1. stability control, which refers to a robot's balance, and
  2. motion control, which refers to a robot's ability to move.

Stability control is particularly difficult for bipedal systems, which must maintain balance in the forward-backward direction even at rest.[1] Some robots, especially toys, solve this problem with large feet, which provide greater stability while reducing mobility. Alternatively, more advanced systems use sensors such as accelerometers or gyroscopes to provide dynamic feedback in a fashion that approximates a human being's balance.[1] Such sensors are also employed for motion control and walking. The complexity of these tasks lends itself to machine learning.[2]

Simple bipedal motion can be approximated by a rolling polygon where the length of each side matches that of a single step. As the step length grows shorter, the number of sides increases and the motion approaches that of a circle. This connects bipedal motion to wheeled motion as a limit of stride length.[2]

Two-legged robots include:

  • Boston Dynamics' Atlas
  • Toy robots such as QRIO and ASIMO.
  • NASA's Valkyrie robot, intended to aid humans on Mars.[11]
  • The ping-pong playing TOPIO robot.

Four-legged

[edit]
Quadruped robot "BigDog" was being developed as a mule that could traverse difficult terrain.

Quadrupedal or four-legged robots exhibit quadrupedal motion. They benefit from increased stability over bipedal robots, especially during movement. At slow speeds, a quadrupedal robot may move only one leg at a time, ensuring a stable tripod. Four-legged robots also benefit from a lower center of gravity than two-legged systems.[1]

Four legged robots include:

  • The TITAN series, developed since the 1980s by the Hirose-Yoneda Laboratory.[1]
  • The dynamically stable BigDog, developed in 2005 by Boston Dynamics, NASA's Jet Propulsion Laboratory, and the Harvard University Concord Field Station.[12]
  • BigDog's successor, the LS3.
  • Spot by Boston Dynamics
  • ANYmal and ANYmal X (the explosion-proof version) by ANYbotics[13]
  • MIT's new back flipping mini Cheetah robot
  • Aliengo[14] by Unitree Robotics
  • Stanford Pupper[15]
  • The Open Dynamic Robot Initiative robots with 8DOF and 12DOF [16][17]
  • Botcat-robot with a moving spine [18][19]
  • Cheetah-Cub robot from the Biorobotics Laboratory [20][21]
  • Oncilla robot from the Biorobotics Laboratory(open source)[22][23]
  • Morti robot from the Dynamic Locomotion Group [24][25]
  • Honey Badger by MAB Robotics[26]

Six-legged

[edit]

Six-legged robots, or hexapods, are motivated by a desire for even greater stability than bipedal or quadrupedal robots. Their final designs often mimic the mechanics of insects, and their gaits may be categorized similarly. These include:

  • Wave gait: the slowest gait, in which pairs of legs move in a "wave" from the back to the front.
  • Tripod gait: a slightly faster step, in which three legs move at once. The remaining three legs provide a stable tripod for the robot.[1]

Six-legged robots include:

  • LAURON, a six-legged, biologically inspired robot being developed at the FZI Forschungszentrum Informatik in Germany.
  • Odex, a 375-pound hexapod developed by Odetics in the 1980s. Odex distinguished itself with its onboard computers, which controlled each leg.[6]
  • Genghis, one of the earliest autonomous six-legged robots, was developed at MIT by Rodney Brooks in the 1980s.[1][27]
  • The modern toy series, Hexbug.

Eight-legged

[edit]

Eight-legged legged robots are inspired by spiders and other arachnids, as well as some underwater walkers. They offer by far the greatest stability, which enabled some early successes with legged robots.[1]

Eight-legged robots include:

Hybrids

[edit]

Some robots use a combination of legs and wheels. This grants a machine the speed and energy efficiency of wheeled locomotion as well as the mobility of legged navigation. Boston Dynamics' Handle, a bipedal robot with wheels on both legs, is one example.[29]

See also

[edit]

References

[edit]
Revisions and contributorsEdit on WikipediaRead on Wikipedia

Legged robot

View on Grokipedia
from Grokipedia
A legged robot is a type of that employs articulated limbs, typically in the form of legs, to achieve locomotion, enabling across uneven, unstructured, or deformable terrains where wheeled or tracked systems often fail. Inspired by biological systems such as animal and movement, these robots offer superior adaptability, stability, and obstacle-traversing capabilities through mechanisms like point contacts and dynamic balancing. Unlike continuous-contact designs, legged robots alternate between stance and swing phases, which enhances energy efficiency on soft ground but introduces complexities in control and . The field of legged robotics traces its origins to the mid-20th century, with foundational studies on stability and emerging in the 1960s, such as early work on quadruped locomotion algorithms. Key milestones include the development of WABOT-1 in 1973, the first full-scale anthropomorphic bipedal robot capable of walking and interacting with its environment, and Marc Raibert's one-legged hopping robot in the 1980s, which pioneered dynamic control strategies for balance during rapid motion. Subsequent advancements, driven by institutions like MIT and , have integrated bio-inspired designs with modern sensors, actuators, and AI, achieving speeds up to 5.76 km/h in quadrupeds like Spot and enabling real-world deployments. Recent developments as of 2025 include AI-enhanced humanoid robots and hybrid wheeled-legged systems for versatile multi-terrain applications. Legged robots are classified by the number of legs and structural configuration, including monopods (one leg) for hopping, bipeds (two legs) for humanoid tasks, quadrupeds (four legs) for versatile mobility akin to dogs, and hexapods (six legs) for stable insect-like traversal. They incorporate varied leg designs, such as telescopic for simple vertical motion or articulated with multiple (DoFs) for biomimetic flexibility, often using series elastic actuators to mimic muscle compliance. Primary applications encompass search-and-rescue in zones, planetary exploration on rough extraterrestrial surfaces, industrial inspections in hazardous areas, and in human-centric or dynamic environments, with over 60 physical prototypes validated for such uses as of 2023. Despite progress, challenges persist in energy efficiency, real-time adaptation to slippery or variable terrains, and scaling for multi-robot coordination.

Fundamentals

Definition and Classification

A legged robot is a type of that employs one or more articulated legs for locomotion, enabling it to traverse complex environments by stepping over obstacles, climbing stairs, and adapting to uneven , in contrast to wheeled or tracked robots which are limited on irregular surfaces. This design mimics biological locomotion systems, such as those in animals, to achieve versatility in mobility. The fundamental advantage lies in the ability to maintain contact with the ground using discrete support points, allowing for negotiation of gaps and rough textures that would immobilize other robotic platforms. Legged robots are classified primarily by the number of legs, which influences stability, speed, and energy efficiency: unipedal (one leg, often for or hopping), bipedal (two legs, emulating walking), quadrupedal (four legs, common for dog-like ), and multi-legged (six or more, inspired by insects for robustness on cluttered ). Additional schemes include types, distinguishing static gaits—where the center of mass remains within the support at all times for inherent stability, as in tripod gaits—and dynamic gaits, which rely on inertial forces and control algorithms for faster but less stable motion, such as trotting or galloping. Autonomy levels further categorize them as teleoperated (-controlled), semi-autonomous (with assisted ), or fully autonomous (self-reliant in and decision-making). Key concepts in legged robot design encompass the (DOF) in joints, typically ranging from 3 to 6 per to replicate natural — for instance, 3 DOF at the (abduction/adduction, flexion/extension, ), 1 at the (flexion/extension), and 2 at the ankle (inversion/eversion, plantarflexion/dorsiflexion). Basic performance metrics include step length (distance per cycle), stride (steps per unit time), and duty factor (proportion of cycle time a is in stance phase), which quantify locomotion efficiency and are used to compare designs across categories. Early classifications in literature, such as those proposed by Shigeo Hirose in the 1980s, differentiated insect-like designs (multi-legged with parallel for load distribution) from mammal-like ones (fewer legs with serial chains for dynamic balance).

Historical Development

The development of legged robots traces its roots to early 20th-century conceptual designs inspired by biological locomotion, with significant precursors emerging in the . In 1940, British engineers A. C. Hutchinson and F. S. Smith proposed the "Walking SuperTank," an early attempt at a multi-legged capable of traversing rough using independent control, marking one of the first serious efforts to walking machines for practical applications. Around the same time, neuroscientist W. Grey Walter's autonomous "" robots, introduced in 1948, served as foundational precursors to mobile ; although wheeled, these vacuum-tube-based machines demonstrated reactive behaviors and environmental adaptation that influenced later legged systems seeking bio-inspired autonomy. By the , advanced to actual legged prototypes, such as the 1961 eight-legged walker developed by Space General Corporation, which navigated outdoor using mechanical linkages, and the 1968 "Phony Pony" quadruped at under Robert B. McGhee, which employed finite-state control for basic walking gaits. The 1970s and 1980s saw breakthroughs in computer-controlled legged mobility, shifting from mechanical to dynamic systems. In 1977, McGhee's hexapod robot at Ohio State achieved wave-gait walking using digital computation, enabling stable locomotion over uneven surfaces. Japanese researcher Shigeo Hirose's TITAN series, starting with TITAN III in the early at , introduced quadrupedal designs capable of climbing stairs and obstacles with simple sensors, emphasizing lightweight structures and passive dynamics for energy efficiency. These efforts laid groundwork for more agile machines, including Marc Raibert's 1983 one-legged hopper at MIT's Leg Laboratory, which balanced and traveled using dynamic control principles that became seminal for subsequent legged platforms. Progress accelerated in the 1990s and 2000s with commercial and challenge-driven innovations focused on humanoid and entertainment applications. Sony's AIBO, released in 1999, was a quadrupedal robot integrating AI for pet-like behaviors, popularizing legged robotics in consumer markets while advancing sensor fusion and autonomy. Honda's ASIMO, unveiled in 2000, represented a milestone in bipedal humanoids, demonstrating stable walking, object manipulation, and human interaction through refined zero-moment point control. The DARPA Grand Challenges of 2004 and 2005, though primarily for wheeled autonomous vehicles, spurred broader advancements in rugged terrain navigation. Separate DARPA initiatives funded legged projects like Boston Dynamics' BigDog in 2005 under Raibert's leadership, a quadruped that trotted dynamically over rough ground using hydraulic actuation. Jerry Pratt's contributions during this era, including virtual model control and series elastic actuators developed at MIT and later IHMC, enhanced force-sensitive compliance in bipedal systems, enabling more robust balance and interaction. From the 2010s onward, legged robots integrated AI and advanced dynamics for real-world deployment, with leading through iterative designs like the 2013 Atlas humanoid, which evolved by 2025 to perform , manipulation, and whole-body control in unstructured environments. The 2016 Spot quadruped further exemplified commercial viability, offering modular payloads for inspection and navigation in industrial settings. Influenced by DARPA's 2012-2015 Robotics Challenge, these platforms emphasized AI-driven adaptation for disaster scenarios. By 2025, trends shifted toward soft-legged robots, such as foldable leg-wheel hybrids and compliant designs, enhancing disaster response by enabling traversal of rubble and narrow spaces without rigid failure risks.

Locomotion Mechanics

Gait Patterns

Gait in legged robots refers to the coordinated sequence of leg movements that produce locomotion, characterized by repeating cycles of stance (foot in contact with the ground) and swing (foot lifted and repositioned) phases for each leg. A complete gait cycle spans the time for all legs to return to their initial configuration relative to the body. Gaits are broadly classified as static or dynamic based on stability mechanisms and speed requirements. Static gaits ensure continuous support with the center of mass projection always within the formed by grounded feet, typically requiring at least three legs in contact for multi-legged systems; these prioritize stability and are suitable for uneven terrain at low speeds. Examples include the ripple gait, where legs lift sequentially from one side to the other, and the tripod gait, alternating between two sets of three non-adjacent legs. Dynamic gaits, in contrast, feature periods of reduced support (fewer than three legs grounded), relying on inertial forces and active control for balance to achieve higher velocities. Representative dynamic patterns are bounding, with paired fore and hind legs on the same side swinging together, and galloping, an asymmetric sequence enabling rapid acceleration. Gait patterns vary by the number of legs to optimize support and efficiency. For bipedal robots, the walk is a static with legs alternating in near 180-degree phase opposition, maintaining double support phases, while the run is dynamic, incorporating a flight phase where both legs are airborne. Quadrupedal patterns encompass the , a static with sequential leg lifts at 90-degree phase intervals (phase difference of 0.25), providing broad support; the , a dynamic synchronizing diagonal legs (phase difference of 0.5) for balanced propulsion; and the pace, pairing ipsilateral legs (phase difference of 0 between laterals). In hexapedal robots, the alternating tripod statically coordinates three legs in support while the others swing, ensuring overlap in stance for stability, whereas the wave propagates leg lifts sequentially across the body (metachronal rhythm) to minimize energy while traversing obstacles. Central parameters govern gait design and performance. The duty factor β=tstanceTcycle\beta = \frac{t_\text{stance}}{T_\text{cycle}}, where tstancet_\text{stance} is the stance duration and TcycleT_\text{cycle} the full cycle time, distinguishes walks (β>0.5\beta > 0.5, overlapping stances) from runs (β<0.5\beta < 0.5, with flight phases); for static quadrupedal s, β≥0.75\beta \geq 0.75 ensures three-leg support. Phase differences ϕ\phi quantify timing offsets between legs relative to the cycle (e.g., ϕ=0.5\phi = 0.5 for trotting diagonals), enabling symmetrical or asymmetrical coordination. The Fr=v2gLFr = \frac{v^2}{gL}, with vv as forward velocity, gg , and LL leg length or hip height, nondimensionalizes speed to predict transitions, such as from walk to near Fr≈0.5Fr \approx 0.5, mirroring biological scaling laws. Many legged gaits draw bio-inspiration from , adapting mammalian symmetrical trots (diagonal leg pairing for efficient energy transfer) or alternating tripods (centralized for robust traversal). The stride length SS, defined as the net displacement of the body center over one gait cycle, approximates S=S = step length ×\times number of legs in support for static multi-legged patterns, where sequential steps cumulatively advance the while maintaining overlap.

Support and Stability

Support patterns in legged robots determine the number of legs in contact with the ground to ensure equilibrium, with static stability achieved when at least three legs form a support for quadrupedal systems, allowing the center of mass (CoM) projection to remain within this to prevent tipping. For multi-legged robots, this expands with more contact points, providing redundancy against perturbations, whereas bipedal configurations rely on dynamic adjustments since only one or two legs contact at a time, reducing the static support area. Stability criteria for dynamic balance in legged robots include the (ZMP), defined as the point on the ground where the net moment of inertial and forces has no horizontal component, ensuring balance. A standard expression for the ZMP location in the x-direction for a multi-link is: xzmp=∑imi(g+z¨i)xi−∑imizix¨i−∑ihiθ¨y,i∑imi(g+z¨i)x_{zmp} = \frac{\sum_i m_i (g + \ddot{z}_i) x_i - \sum_i m_i z_i \ddot{x}_i - \sum_i h_i \ddot{\theta}_{y,i} }{\sum_i m_i (g + \ddot{z}_i)} where mim_i is the of link ii, x¨i\ddot{x}_i and z¨i\ddot{z}_i are its horizontal and vertical accelerations, xix_i and ziz_i are its horizontal position and height, gg is , hih_i is the height of the CoM for angular terms, and θ¨y,i\ddot{\theta}_{y,i} is the around the y-axis for link ii; maintaining the ZMP within the support polygon guarantees stability during motion. Complementing ZMP, the capture point serves as a predictive metric for balance recovery, representing the point on the ground to which the robot must step to halt divergent motion under linear dynamics, enabling proactive foothold planning. Factors affecting stability encompass terrain unevenness, which alters the support polygon and induces CoM shifts, payload variations that redistribute mass and torque, and sensor feedback loops that enable real-time CoM adjustments via force or inertial measurements. Margins of stability employ Lyapunov-like metrics to quantify gait robustness, assessing the basin of attraction around equilibrium states to predict recovery from disturbances without falling. In rough environments, these metrics facilitate tip-over avoidance by optimizing leg placements to maximize the minimum distance from the CoM projection to polygon edges, as demonstrated in quadrupedal traversal where stability margins exceed 0.1 m to handle slopes up to 30 degrees.

Design and Components

Actuators and Sensors

Legged robots rely on actuators to generate the forces and motions required for locomotion, with electric motors being a prevalent choice due to their precision and ease of control. (DC) motors and servo motors provide accurate position and velocity control through feedback mechanisms, enabling fine adjustments in trajectories essential for stable walking. Hydraulic actuators offer superior and force output, making them suitable for dynamic, high-torque applications in rough . For instance, the quadruped robot employs hydraulic cylinders to achieve robust leg movements capable of supporting its 109 kg mass while traversing uneven surfaces at speeds up to 2.7 m/s. Pneumatic actuators, often used in , provide compliant and lightweight actuation through pressurized air, facilitating adaptive deformation for irregular environments. These systems excel in bio-inspired designs where flexibility is prioritized over rigidity. Series elastic actuators (SEAs) incorporate a spring element between the motor and load to enable compliant force control, where the output force is determined by spring deflection according to F=kΔxF = k \Delta x, with kk as the spring stiffness and Δx\Delta x as the deflection. This design enhances shock absorption and energy efficiency during impacts, as demonstrated in early legged prototypes. Recent advancements in the have seen a shift toward high-torque electric actuators and quasi-direct drive systems, as in ' electric introduced in 2024, which replaces hydraulic systems for improved energy efficiency, reduced weight, and quieter operation while maintaining dynamic performance. Sensors in legged robots are categorized into proprioceptive types, which monitor internal states, and exteroceptive types, which perceive the external environment. Proprioceptive sensors include joint encoders for measuring angular positions and inertial measurement units () for tracking body orientation and , providing essential data for balance maintenance. Exteroceptive sensors, such as force/torque sensors embedded at the feet, detect ground reaction forces to inform foot placement, while and cameras enable terrain mapping and obstacle avoidance by generating 3D environmental models. Actuator-sensor integration often involves torque control loops using proportional-integral-derivative (PID) algorithms, expressed as u(t)=Kpe(t)+Ki∫e(t) dt+Kde˙(t)u(t) = K_p e(t) + K_i \int e(t) \, dt + K_d \dot{e}(t), where e(t)e(t) is the error, and Kp,Ki,KdK_p, K_i, K_d are tuning parameters; this feedback ensures precise force application based on inputs. Energy efficiency trade-offs arise in actuator selection, as hydraulic systems deliver higher power density (up to 1 kW/kg) but consume more energy overall compared to electric motors (around 0.5 kW/kg), influencing design choices for battery-powered versus tethered operations. Recent advancements in the 2020s include soft actuators based on , which deform under to mimic biological muscle flexibility, offering high strain (up to 100%) and low weight for bio-mimetic legged structures.

Control Systems

Control systems in legged robots encompass the algorithmic frameworks that coordinate multi-joint movements, ensure dynamic balance, and enable adaptive locomotion across varied terrains. These systems integrate sensing data with computational models to generate precise motor commands, often operating in real-time to mimic biological locomotion while handling the underactuated nature of legged platforms. Seminal work in this domain, such as the development of hierarchical architectures in the early 2000s, has allowed robots to transition from rigid, pre-programmed walks to robust, environment-responsive gaits. A core approach is hierarchical control, which decomposes the problem into layers for efficiency. At the low level, joint torques are computed using to achieve desired end-effector positions, formulated as q=f(x)\mathbf{q} = f(\mathbf{x}), where q\mathbf{q} represents joint angles and x\mathbf{x} the Cartesian pose; this enables precise foot placement derived from higher-level goals. Mid-level control generates rhythmic patterns via (CPGs), modeled by coupled oscillators such as θ˙i=ω+∑Ksin⁡(θj−θi)\dot{\theta}_i = \omega + \sum K \sin(\theta_j - \theta_i), where θi\theta_i is the phase of oscillator ii, ω\omega the base , and KK the coupling strength, facilitating stable gait cycles like trotting or galloping. High-level planning employs (SLAM) for path optimization, integrating environmental data to select feasible trajectories while avoiding obstacles. Key techniques include model predictive control (MPC), which optimizes future states over a horizon to minimize a cost function like min⁡J=∑(xk−xref)2+uk2\min J = \sum ( \mathbf{x}_k - \mathbf{x}_{ref} )^2 + \mathbf{u}_k^2 , balancing tracking accuracy and control effort for terrain adaptation; this has been pivotal in enabling energy-efficient locomotion on uneven surfaces. Reinforcement learning (RL) has emerged for adaptive gait synthesis, particularly in the 2020s, with applications like Boston Dynamics' Atlas robot using RL policies trained via simulation to perform complex parkour maneuvers and, as of 2025, diverse locomotion modes including walking, running, and crawling with references from human motion capture, achieving zero-shot generalization to real-world dynamics. These methods address the high dimensionality of legged systems by learning policies that recover from perturbations, outperforming traditional controllers in robustness. Challenges in these systems revolve around real-time computation, requiring control loops at frequencies like 100 Hz to match biological reflexes, often achieved through optimized solvers and . is another focus, with algorithms detecting leg failures and redistributing loads via online reconfiguration, ensuring continued operation despite hardware limitations. The of control systems traces from rule-based methods in the , reliant on explicit kinematic models for simple terrains, to post-2015 learning-based paradigms incorporating deep neural networks for end-to-end . By 2025, integrations of neural networks enable zero-shot to novel environments, as demonstrated in RL frameworks that synthesize without retraining, marking a shift toward biologically inspired, scalable .

Types and Examples

Bipedal Robots

Bipedal robots, characterized by their two-legged locomotion, mimic human upright mobility and typically feature high exceeding 20 in total to enable complex movements such as arm swinging and torso adjustments for balance. This design allows for versatile interaction with human environments but introduces significant challenges due to the narrow support base formed by the feet, which necessitates precise control to prevent tipping. To maintain stability, these robots often rely on the (ZMP) criterion, originally formulated by Vukobratović, which requires active ankle torque adjustments to keep the projection of the center of mass within the support polygon. Energy efficiency in bipedal walking is further enhanced through passive dynamics, as exemplified by the compass gait model, where gravitational and inertial forces drive natural, limit-cycle motion on slopes without continuous actuation. Notable examples include Honda's , unveiled in 2000, which achieved stable bipedal walking at speeds up to 2.7 km/h using a combination of zero-moment point control and pattern generation for smooth transitions. Another landmark is ' Atlas, first developed in 2013 and iterated through 2025, standing approximately 1.8 m tall and weighing 150 kg in its early hydraulic version, capable of dynamic maneuvers such as backflips and elements through advanced and whole-body control. These robots often incorporate bipedal patterns like heel-to-toe walking to optimize energy use and terrain adaptability. In the 2020s, advancements have extended bipedal principles to prosthetic integrations, such as Össur's Power Knee, a microprocessor-controlled device that provides active torque assistance for amputees, improving sit-to-stand transitions and walking symmetry as demonstrated in clinical studies pairing it with custom algorithms. Similarly, Tesla's Optimus prototype, revealed in 2022, leverages AI for autonomous bipedal tasks like object manipulation and navigation, aiming to perform repetitive human labor with integrated vision and planning. These developments highlight ongoing efforts to address balance challenges while expanding practical humanoid applications.

Quadrupedal Robots

Quadrupedal robots, featuring four legs arranged in a mammalian-inspired configuration, offer enhanced stability compared to bipedal designs by providing a wider support polygon that enables static balance even on uneven . This geometric advantage allows the robot's to remain within the formed by the leg contact points, reducing the need for complex dynamic balancing and facilitating reliable locomotion over rough surfaces such as rocks or slopes. Unlike bipedal robots, which rely on active control to maintain equilibrium with only two points of contact, quadrupeds can employ simpler control strategies for traversal in challenging environments. Key design features of quadrupedal robots include articulated legs capable of trots and gallops, achieving speeds up to 5 m/s or more, as demonstrated by the MIT Cheetah in 2013, which sprinted at approximately 22 km/h (6 m/s) using proprioceptive actuators for high-dynamic running. These robots often incorporate modular components for adaptability, with examples like , introduced in 2016 and updated through 2025, weighing 25 kg and supporting modular payloads for tasks such as terrain inspection. Similarly, the ANYmal robot from , developed in 2016, utilizes hydraulic actuators to enable dynamic maneuvers, including jumps up to 20 cm in height, enhancing its capability to navigate obstacles in rugged settings. Spot employs electric actuators for efficient operation, while ANYmal's hydraulic systems provide high force output for demanding jumps. Drawing bio-inspiration from canine locomotion, quadrupedal robots replicate gaits such as the , where diagonal leg pairs exhibit a 0.5 phase lag to ensure smooth and stability. This coordination mimics natural animal patterns, allowing efficient energy transfer during movement over irregular terrain. Recent advancements in have extended these bio-inspired designs to swarm capabilities, enabling multi-robot coordination for collective tasks like distributed exploration in complex environments. Performance metrics highlight their practicality, with Spot offering a payload capacity of up to 14 kg and a battery life of approximately 90 minutes for sustained operations.

Multi-Legged Robots

Multi-legged robots, featuring six or more , excel in environments requiring high and efficient traversal of irregular terrains due to their inherent . This design allows continued operation despite mechanical failures, such as leg loss, which is critical for applications like planetary exploration or hazardous inspections. For instance, in hexapod configurations, wave gaits enable sustained mobility by redistributing load among remaining legs, preserving static stability even with one or two legs compromised. A primary advantage lies in the ability to employ duty factors exceeding 0.5, where the fraction of the gait cycle during which a leg remains in stance phase ensures at least three legs (and often more) are always supporting the body. This configuration guarantees perpetual ground contact, minimizing the risk of instability on uneven surfaces and enhancing overall efficiency compared to designs with fewer legs. Gait patterns in multi-legged robots are tailored to leg count for optimal balance and speed. Hexapods commonly utilize the alternating tripod gait, in which legs on opposite sides alternate support—three legs bear the weight while the other three swing forward—facilitating rapid, stable progression. Octopods, by contrast, often employ radial sequences, coordinating leg movements in a spoke-like pattern around the body to distribute forces evenly and adapt to obstacles. Scalability in these systems follows an approximate rule where the number of support legs is about n/2 for a robot with n legs in dynamic yet stable gaits, allowing larger leg counts to handle greater payloads without proportional increases in complexity. Prominent examples include NASA's (All-Terrain Hex-Limbed Extra-Terrestrial Explorer), a six-legged developed in the late 2000s and 2010s for lunar and planetary mobility, capable of traversing rough terrain while transporting heavy habitats or rovers using its versatile limbs as both legs and manipulators. In the realm of explosive ordnance disposal (EOD), early 2000s prototypes explored multi-legged designs for enhanced maneuverability in debris-strewn areas, building on tracked systems like iRobot's to improve in urban combat zones. More recently, 2024 advancements feature hexapod s optimized for lunar , such as compact designs with adaptive gaits for low-gravity surfaces, demonstrating improved energy efficiency and obstacle negotiation. Emerging trends in 2025 emphasize bio-hybrid multi-legged micro-robots, integrating living insect legs or muscle tissues with synthetic frames to achieve lifelike agility at scales of 1-5 cm. These designs leverage biological actuators for superior power-to-weight ratios and adaptability, targeting applications in confined or dynamic environments like swarm-based planetary scouting.

Hybrid Designs

Hybrid designs in legged robots integrate legs with alternative locomotion mechanisms, such as wheels or tracks, to enhance versatility across diverse terrains by leveraging the strengths of each mode. Leg-wheel hybrids typically employ wheels at the end of legs for efficient traversal on flat surfaces while allowing legs to negotiate obstacles, enabling seamless transitions between rolling and stepping. Leg-track hybrids, in contrast, combine legs with continuous tracks to improve performance on soft or deformable terrains, where tracks provide better traction and legs offer elevated mobility for climbing or stepping over irregularities. Prominent examples include the ANYmal quadruped from 's Robotic Systems Lab, which in the 2020s was adapted with torque-controlled wheels for switchable hybrid modes, allowing dynamic walking-driving gaits in environments like underground tunnels during the Subterranean Challenge. NASA's DuAxel, introduced in 2020 and prototyped through 2022, features detachable wheeled legs based on the Axel rover design, enabling a four-wheeled base for planetary traversal that can undock into independent single-axle units to rappel steep slopes or explore craters on the or Mars. For urban applications in 2025, the LEVA wheeled-legged robot from demonstrates agile navigation in city environments, using to achieve energy-efficient locomotion on sidewalks and stairs. These designs introduce trade-offs, including increased mechanical complexity from additional (DOF) in wheel or track actuators, which can raise control challenges and overhead compared to pure legged systems. However, they yield significant gains, such as up to twice the speed on flat versus purely legged locomotion, by minimizing -intensive leg swinging during rolling phases. Control integration relies on mode-switching algorithms that detect variations—via sensors like , IMUs, or force feedback—to trigger transitions, often using or hybrid automata to optimize trajectories in real-time. For instance, online in wheeled-legged systems separates wheel and body planning to ensure robust, disturbance-resistant switching. In unstructured settings, integrated path planning and mode decision algorithms, such as IPP-MD, fuse estimation with locomotion policies to select wheel, , or hybrid modes autonomously.

Applications and Challenges

Practical Uses

Legged robots have found practical applications in operations, where their ability to navigate unstable rubble and debris provides critical access to hazardous areas. For instance, ' Spot robot has been deployed in scenarios to inspect collapsed structures and locate survivors, leveraging its mobility to traverse uneven terrain without endangering human teams. Similarly, ANYbotics' ANYmal has been used in simulated disaster sites for autonomous exploration and data collection in environments mimicking earthquake rubble. In industrial inspection, legged robots excel in monitoring hard-to-reach infrastructure in high-risk settings such as and nuclear facilities. ANYmal robots are routinely employed for gas leak detection and structural assessments on offshore oil platforms, enduring extreme weather while reducing human exposure to toxic or explosive hazards. In nuclear plants, like Finland's Onkalo repository, these robots conduct remote inspections of radiation-prone areas, using advanced sensors to map and analyze conditions autonomously. Agriculture benefits from quadrupedal robots for precise monitoring across uneven fields, where wheeled alternatives often struggle. Models like Unitree's quadrupeds equip farmers with AI-driven vision systems to track growth, , and pest detection, enabling data collection over varied landscapes without —as demonstrated in partnerships for smart farming applications as of July 2025. Compared to wheeled robots, legged designs offer superior obstacle clearance and terrain versatility, climbing steps up to 30 cm high—such as those navigated by Spot—while maintaining stability on slopes up to ±30 degrees and surfaces like sand or stairs. This adaptability allows 40-50% greater negotiation of irregular obstacles, as demonstrated in hybrid wheel-legged systems that outperform pure wheeled platforms in rough environments by dynamically adjusting leg configurations for elevation changes. Notable case studies highlight commercial adoption: Boston Dynamics' Spot robots have been deployed in warehouse facilities by retailers like the Otto Group for inventory audits and maintenance in dynamic logistics spaces across 20 facilities as of 2024. In elderly care, humanoid prototypes such as MIT's E-BAR robot assist with lifting tasks, supporting users up to full body weight during transfers from beds to wheelchairs, thereby reducing caregiver strain in home settings. The economic impact underscores growing viability, with the global legged robot market valued at approximately $0.6 billion in 2023 and projected to reach $1.5 billion by 2030, driven by demand in and sectors.

Limitations and Future Directions

Legged robots face significant limitations in energy efficiency, with studies indicating that they consume substantially more power per unit distance traveled compared to wheeled counterparts, often due to the dynamic balancing and actuation required for locomotion. For instance, optimal parameters in legged systems can reduce energy use, but predefined gaits frequently fail to adapt to varying speeds or loads, exacerbating inefficiency. Mechanical complexity further compounds these issues, as the numerous joints and actuators increase vulnerability to ; highlights that legged designs exhibit higher failure rates in rugged environments, with obstacle avoidance success rates typically ranging from 90% to 95%, implying a 5-10% failure margin that is notably elevated relative to simpler wheeled systems. Additionally, high and costs, often exceeding $50,000 per unit for commercial models like quadrupedal platforms, limit widespread adoption. Key challenges include limited battery life, particularly for dynamic operations, where most legged robots operate for under two hours on a single charge before requiring recharging or swapping. Scalability for swarm deployments remains problematic, as coordinating large groups of legged robots demands robust decentralized control to manage communication overhead and collision avoidance without centralized processing, which current systems struggle to achieve efficiently. Ethical concerns are also prominent, especially in military applications; in 2025, UN discussions emphasized the risks of autonomous weapons systems, including AI-driven robots, urging prohibitions on those lacking meaningful human control to prevent unintended escalations in warfare. Looking to future directions, advancements in , such as multimodal large language models integrated for task planning, promise to enhance legged robot autonomy by enabling and environmental reasoning in real-time scenarios as of 2025. Soft robotics innovations, including muscle-like actuators, offer potential efficiency gains of up to 50% through bio-inspired designs that mimic natural contraction for reduced energy loss in locomotion. Neuromorphic computing emerges as a pathway for low-power control, emulating brain-like processing to minimize computational demands in resource-constrained legged systems. Projections suggest significant growth in legged robots by 2035, with the market estimated to reach $38 billion and millions of units deployed in industrial and domestic settings to handle repetitive tasks. Hybrid bio-robots, exemplified by cyborgs integrating living tissues with for , represent another frontier, leveraging biological efficiency for applications in search-and-rescue where traditional legged designs falter.

References

  1. https://pmc.ncbi.nlm.nih.gov/articles/PMC11639468/
  2. https://www.frontiersin.org/journals/mechanical-engineering/articles/10.3389/fmech.2023.1142421/full
  3. https://spj.science.org/doi/10.34133/cbsystems.0365
  4. https://humanoidroboticstechnology.com/articles/top-12-humanoid-robots-of-2025/
  5. https://cyberneticzoo.com/walking-machines/1940-walking-supertank-hutchinson-and-smith-british/
  6. https://home.csulb.edu/~wmartinz/content/w-grey-walter-and-his-turtle-robots.html
  7. http://www.ai.mit.edu/projects/leglab/old-leglab/background/history.html
  8. https://www.sciencedirect.com/science/article/pii/S2090447920302501
  9. https://ifr.org/robot-history
  10. https://journals.sagepub.com/doi/10.1177/1077546307078276
  11. https://dspace.mit.edu/handle/1721.1/7082
  12. https://www.researchgate.net/publication/252639079_Series_Elastic_Actuators_for_legged_robots
  13. https://www.darpa.mil/about/innovation-timeline
  14. https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202512435
  15. https://www.impactlab.com/2025/04/25/flexible-soft-robots-poised-to-transform-disaster-response-and-medicine/
  16. https://www.atlantis-press.com/article/25886044.pdf
  17. https://www.sciencedirect.com/science/article/pii/S2667379721000292
  18. https://www.ri.cmu.edu/pub_files/pub2/wettergreen_david_1992_1/wettergreen_david_1992_1.pdf
  19. https://eprints.whiterose.ac.uk/id/eprint/201823/1/bio_gait.pdf
  20. https://pmc.ncbi.nlm.nih.gov/articles/PMC5321742/
  21. https://cjme.springeropen.com/articles/10.1186/s10033-019-0321-2
  22. https://pmc.ncbi.nlm.nih.gov/articles/PMC7506578/
  23. https://lall.stanford.edu/papers/bretl_eqmcut_ieee_tro_projection_2008_08_01_01/pubdata/entry.pdf
  24. https://www.cs.cmu.edu/~cga/legs/vukobratovic.pdf
  25. https://scaron.info/robotics/zero-tilting-moment-point.html
  26. http://www.ambarish.com/paper/Pratt_Goswami_Humanoids2006.pdf
  27. https://journals.sagepub.com/doi/10.5772/52894
  28. https://hal.science/hal-03361804/file/ICRA_2021.pdf
  29. https://groups.csail.mit.edu/robotics-center/public_papers/Wieber15.pdf
  30. https://www.researchgate.net/publication/287716986_Lyapunov_Stability_Margins_for_humanoid_robot_balancing
  31. https://www.tandfonline.com/doi/full/10.1080/01691864.2018.1455606
  32. https://www.cs.swarthmore.edu/~meeden/DevelopmentalRobotics/bigdog.pdf
  33. https://ieeexplore.ieee.org/document/7090302
  34. https://www.cs.cmu.edu/~cga/legs/jh1c.pdf
  35. https://bostondynamics.com/blog/an-electric-new-era-for-atlas/
  36. https://academic.oup.com/nsr/article/10/5/nwac214/6762939
  37. https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2020.510757/full
  38. https://www.youtube.com/watch?v=I44_zbEwz_w
  39. https://www.romela.org/wp-content/uploads/2015/05/Constrained-Analytical-Trajectory-Filter-for-stabilizing-humanoid-robot-motions.pdf
  40. https://www.cambridge.org/core/journals/robotica/article/multilevel-control-of-zeromoment-pointbased-humanoid-biped-robots-a-review/93E028820A47A97C242A9027A6CC4388
  41. http://ruina.tam.cornell.edu/research/topics/locomotion_and_robotics/efficiency_speed_scaling/efficiency_speed_scaling.pdf
  42. https://robots.ieee.org/robots/asimo/
  43. https://bostondynamics.com/blog/leaps-bounds-and-backflips/
  44. https://www.ossur.com/en-us/prosthetics/knees/power-knee
  45. https://www.tesla.com/AI
  46. https://crl.ethz.ch/papers/ICRA2013.pdf
  47. https://dspace.mit.edu/bitstream/handle/1721.1/85490/870998828-MIT.pdf?sequence=2&isAllowed=y
  48. https://robotsguide.com/robots/spot
  49. https://www.researchgate.net/publication/311758153_ANYmal_-_a_highly_mobile_and_dynamic_quadrupedal_robot
  50. https://www.nature.com/articles/s41467-025-61896-7
  51. https://support.bostondynamics.com/s/article/Spot-Specifications-49916
  52. https://www.researchgate.net/publication/221907923_Locomotion_Analysis_of_Hexapod_Robot
  53. https://pmc.ncbi.nlm.nih.gov/articles/PMC5751548/
  54. https://www.research.ed.ac.uk/files/40020480/Accepted_version_2.pdf
  55. https://journals.sagepub.com/doi/10.5772/10487
  56. https://www.sciencedirect.com/science/article/abs/pii/S0957415811001115
  57. https://www.army-technology.com/features/featuredetect-and-diffuse-the-top-5-military-robots-for-explosive-ordnance-disposal-4372678/
  58. https://www.mdpi.com/2076-3417/14/1/260
  59. https://advanced.onlinelibrary.wiley.com/doi/full/10.1002/aisy.202401105
  60. https://arxiv.org/abs/1909.07193
  61. https://www.mdpi.com/2075-1702/9/1/10
  62. https://rsl.ethz.ch/robots-media/anymal-wheels.html
  63. https://www.jpl.nasa.gov/news/this-transforming-rover-can-explore-the-toughest-terrain/
  64. https://arxiv.org/abs/2503.10028
  65. https://bostondynamics.com/blog/spot-to-the-rescue/
  66. https://www.youtube.com/watch?v=_R29DrAx0Xs
  67. https://www.anybotics.com/robotics/anymal-x/
  68. https://www.anybotics.com/news/autonomous-inspection-for-enhanced-nuclear-safety-at-onkalo/
  69. https://igrownews.com/unitree-robotics-latest-news/
  70. https://www.prnewswire.com/news-releases/from-streets-to-fields-unitrees-go2-steps-into-smart-farming-302513086.html
  71. https://bostondynamics.com/products/spot/
  72. https://pmc.ncbi.nlm.nih.gov/articles/PMC8640219/
  73. https://www.therobotreport.com/otto-group-deploying-boston-dynamics-robots-in-20-facilities/
  74. https://news.mit.edu/2025/eldercare-robot-helps-people-sit-stand-catches-them-fall-0513
  75. https://www.qyresearch.in/report-details/7819845/machinery-equipment-global-legged-robot-market-insights-industry-share-sales-projections-and-demand-outlook-2024-2030
  76. http://meetings.unoda.org/ccw/convention-on-certain-conventional-weapons-group-of-governmental-experts-on-lethal-autonomous-weapons-systems-2025
  77. https://www.goldmansachs.com/insights/articles/the-global-market-for-robots-could-reach-38-billion-by-2035
Add your contribution
Related Hubs
User Avatar
No comments yet.