X2-N: The Robot That Walks, Rolls, and Adapts

by

Author: Denis Avetisyan

Researchers unveil a transformable humanoid robot capable of seamlessly switching between wheeled and legged locomotion for enhanced efficiency and versatility.

The system demonstrates a capacity for morphological reconfiguration, transitioning between ambulation via legs and wheels, and further exhibits the ability to reverse this process, suggesting an inherent adaptability built upon reciprocal structural modes.
The system demonstrates a capacity for morphological reconfiguration, transitioning between ambulation via legs and wheels, and further exhibits the ability to reverse this process, suggesting an inherent adaptability built upon reciprocal structural modes.

This paper details the design, control framework, and experimental validation of X2-N, a dual-mode wheel-legged robot demonstrating improved loco-manipulation capabilities through reinforcement learning and hybrid locomotion strategies.

Conventional wheel-legged robots often compromise between the efficiency of wheeled motion and the adaptability of legged locomotion, typically exhibiting limited stability or manipulation capabilities. This work presents ‘X2-N: A Transformable Wheel-legged Humanoid Robot with Dual-mode Locomotion and Manipulation’, introducing a novel high-degree-of-freedom platform capable of seamless transitions between humanoid and wheel-legged configurations. Through a reinforcement learning-based whole-body control framework, X2-N achieves unified control across hybrid locomotion, transformation, and manipulation tasks, demonstrating strong terrain adaptability and stable loco-manipulation performance. Could this integrated approach pave the way for more versatile and robust robots capable of navigating complex real-world environments?

The Inevitable Failure of Static Design

Conventional robotic locomotion, reliant on either wheels or legs, frequently falters when confronted with the irregularities of real-world environments. Wheeled robots, while efficient on smooth surfaces, become immobilized by obstacles exceeding their ground clearance or by uneven terrain. Conversely, legged robots, though capable of traversing complex landscapes, often sacrifice speed and energy efficiency due to the intricate coordination required for stable gait. These limitations stem from the fundamental rigidity of designs optimized for specific conditions; a robot engineered for flat floors struggles in a rubble-strewn disaster zone, and one built for rocky hills proves unwieldy indoors. This inherent inflexibility highlights the need for robotic systems capable of dynamically adapting their locomotion strategy to overcome environmental challenges and maintain robust performance across a diverse range of terrains.

Current robotic designs frequently prioritize either efficient locomotion or precise manipulation, creating a significant performance bottleneck when both are required in real-world scenarios. Many robots struggle to maintain stability and speed while simultaneously performing intricate tasks, particularly when operating on uneven or unpredictable terrain. This limitation stems from the difficulty of coordinating complex movements across multiple degrees of freedom and maintaining balance during dynamic manipulation. Consequently, existing systems often exhibit reduced efficiency, increased energy consumption, and a diminished capacity to operate autonomously in complex environments – highlighting the need for integrated designs that seamlessly blend robust mobility with dexterous manipulation capabilities.

Current robotic designs frequently prioritize either efficient wheeled movement on smooth surfaces or the adaptable, though often slower, traversal offered by legged systems. However, a significant leap forward necessitates integrating the benefits of both approaches. Researchers are actively developing hybrid locomotion systems – robots capable of seamlessly transitioning between wheeled and legged gaits – to achieve superior robustness and versatility. These designs aim to combine the speed and energy efficiency of wheels for covering long distances on predictable terrain with the ability to step over obstacles, navigate uneven ground, and even perform complex manipulations, much like a human can walk, climb, and reach. This integration promises robots that are not limited by environmental complexity, opening doors to applications in search and rescue, disaster response, and truly autonomous exploration of challenging landscapes.

A variety of classic legged robot platforms demonstrate the diversity of early approaches to bipedal and quadrupedal locomotion.
A variety of classic legged robot platforms demonstrate the diversity of early approaches to bipedal and quadrupedal locomotion.

X2-N: A Temporary Reprieve from Inevitable Obsolescence

The X2-N robot utilizes a transformable leg design allowing it to switch between wheel-contact and foot-contact modes for varied terrain traversal. Each leg incorporates a multi-link structure with integrated actuators enabling both rotary motion for wheel-like rolling and reciprocal motion for traditional walking gaits. This is achieved through a variable geometry configuration where the foot can be retracted and aligned with the leg’s rotational axis, effectively creating a wheel, or extended downwards for foot-based locomotion. The transition between modes is performed dynamically, allowing the robot to adapt its locomotion style based on surface conditions and operational requirements without requiring a complete structural reconfiguration.

The X2-N robot’s design incorporates the Joint Reuse Principle, a methodology focused on minimizing the total number of actuators required for complex movements. This is achieved by strategically configuring joints to serve multiple kinematic functions across both wheel and leg modes. Rather than dedicating specific actuators solely to wheel rotation or leg extension, the system leverages shared actuators and linkages. This reduces overall weight, power consumption, and control complexity while maintaining a sufficient range of motion for varied terrain. By intelligently sharing mechanical components, the X2-N effectively minimizes redundant degrees of freedom, resulting in a more efficient and compact robotic system.

The X2-N robot employs a Locking Mechanism to facilitate stable transitions and operation in both wheel and leg modes. This mechanism utilizes high-torque servo motors and a multi-stage locking system at each wheel-leg joint, rigidly fixing the wheel in line with the leg segment for efficient rolling locomotion. Conversely, the mechanism releases to allow full articulation for foot-contact walking and obstacle negotiation. Rigorous testing demonstrates the Locking Mechanism maintains structural integrity under loads exceeding 1.5x the robot’s weight and exhibits a latching/unlatching time of less than 300 milliseconds, crucial for dynamic mode switching and maintaining traversal efficiency across varied terrains.

The X2-N humanoid demonstrates an expanded workspace and improved manipulability compared to conventional humanoid robots due to its leg-joint configuration.
The X2-N humanoid demonstrates an expanded workspace and improved manipulability compared to conventional humanoid robots due to its leg-joint configuration.

Control: A Brief Delay of the Entropy

The X2-N robot utilizes a Unified Control Framework designed to synergistically combine the strengths of both Model-Based Control and Reinforcement Learning (RL). This framework does not operate as a sequential process, but rather integrates both methodologies within a single control architecture. Model-Based Control provides a predictive and reliable foundation for tasks requiring precision and stability, while RL algorithms are implemented to enhance adaptability and optimize performance in dynamic and unpredictable environments. This integration allows X2-N to leverage the benefits of both approaches – the accuracy of model prediction and the robustness gained through learned behaviors – within a cohesive control system.

Reinforcement Learning (RL) is implemented to develop the locomotion controller for X2-N, allowing the robot to learn optimal movement policies through interaction with simulated and real-world environments. This training process focuses on maximizing reward signals related to speed, stability, and energy efficiency across diverse terrains. The RL-based controller enables agile locomotion, including navigating uneven surfaces, recovering from disturbances, and adapting to changing environmental conditions without explicit programming for each scenario. This adaptive capability is achieved through algorithms that iteratively refine the control policy based on experience, resulting in robust and versatile movement skills.

Model-based control within X2-N utilizes a pre-defined dynamic model of the robot to calculate actuator commands, ensuring accurate tracking of desired trajectories and maintaining stability during complex movements. This approach is critical for dexterous manipulation as it allows for precise positioning of end-effectors and consistent force application, even in the presence of external disturbances or uncertainties in the environment. The reliance on a known model enables predictable behavior and facilitates the execution of intricate manipulation tasks requiring high levels of precision and repeatability, such as assembly, tool use, and delicate object handling.

X2-N employs a control architecture integrating perception, planning, and control to navigate and interact with its environment.
X2-N employs a control architecture integrating perception, planning, and control to navigate and interact with its environment.

Performance: A Temporary Stay Against the Inevitable

The X2-N robot distinguishes itself through a design prioritizing adaptability, achieved via high-degree-of-freedom legs and interchangeable arm configurations. This approach moves beyond the limitations of robots tailored to specific tasks; the high-DoF legs enable nuanced movement and stability across varied terrains, while the modular arm design facilitates quick reconfiguration for different operational demands. This isn’t simply about adding more joints; it’s a deliberate engineering choice to broaden the scope of potential applications, allowing the X2-N to seamlessly transition between tasks like inspection, manipulation, and even locomotion in challenging environments. Consequently, the robot’s versatility significantly reduces the need for specialized hardware or extensive reprogramming, positioning it as a highly efficient platform for a wide spectrum of robotic endeavors.

Rigorous workspace analysis was central to X2-N’s design, leveraging [latex]Yoshikawa’s Manipulability Index[/latex] to meticulously map its range of motion and dexterity. This metric assesses the robot’s ability to move and manipulate objects within its reachable space, effectively quantifying its kinematic performance. The analysis revealed that, despite its unique hybrid locomotion system, X2-N achieves workspace and manipulability characteristics directly comparable to those of conventional humanoid robots with dedicated legs. This ensures the robot isn’t compromised in its ability to perform complex tasks requiring precise movements and manipulation, validating the design’s effectiveness in bridging wheeled and legged mobility without sacrificing dexterity.

X2-N’s capacity to navigate challenging terrains stems from a seamless integration of adaptable design and a dual-mode operational system. This robotic platform isn’t limited to conventional wheeled movement or bipedal locomotion; it swiftly transitions between the two, completing transformations in a remarkable one second. This rapid adaptation is achieved through optimized kinematic performance, informed by detailed workspace analysis, and allows X2-N to maintain operational efficiency whether traversing smooth surfaces with wheels or negotiating uneven ground on foot. The resulting agility enables the robot to respond dynamically to environmental changes, bypassing obstacles and maintaining stability in complex settings where a fixed mode of locomotion would prove restrictive.

The X2-N robot demonstrates versatile locomotion on complex terrains like stairs and slopes using either 4 or 7 degrees of freedom in its arms.
The X2-N robot demonstrates versatile locomotion on complex terrains like stairs and slopes using either 4 or 7 degrees of freedom in its arms.

The pursuit of seamless transitions in X2-N, moving between wheeled and legged locomotion, echoes a fundamental truth about complex systems. It’s a dance not of perfect engineering, but of managed compromise. As Robert Tarjan observed, “The most effective algorithms are often the simplest.” While X2-N isn’t an algorithm, the principle applies; the robot’s transformable design acknowledges that a single, optimal solution is elusive. Instead, the system embraces a spectrum of capabilities, sacrificing absolute efficiency in one mode for adaptability in another. This reflects an understanding that scalability isn’t about achieving perfection, but about gracefully accommodating inevitable limitations – a prophecy of future failure, willingly accepted.

What Lies Ahead?

X2-N’s demonstration of integrated locomotion and manipulation, while a step forward, merely clarifies the scope of what remains unknown. The system’s adaptability isn’t inherent robustness, but a temporary reprieve from the inevitable mismatch between model and world. Each seamless mode transition is, in effect, a localized postponement of systemic failure-a prophecy of the situations where the unified control framework will inevitably falter. Monitoring these failures isn’t about preventing incidents; it’s the art of fearing consciously.

Future work isn’t about perfecting the architecture, but about embracing the emergent properties of imperfect systems. True resilience begins where certainty ends. The focus should shift from achieving seamless transitions to designing for graceful degradation-for systems that anticipate their own limitations and reconfigure around them. The challenge isn’t loco-manipulation itself, but the creation of robots that can negotiate the unpredictable terrain of their own incomplete knowledge.

Ultimately, X2-N and its successors will not be judged by what they can do, but by what they reveal about the limits of control. That’s not a bug-it’s a revelation. The path forward lies not in building more complex systems, but in cultivating the capacity to understand, and even appreciate, the beautiful, chaotic unfolding of their inevitable failures.

Original article: https://arxiv.org/pdf/2604.21541.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-04-24 15:04