pith. sign in

arxiv: 2507.22345 · v2 · submitted 2025-07-30 · 💻 cs.RO

A Reconfigured Wheel-Legged Robot for Enhanced Steering and Adaptability

Zhicheng Song , Jinglan Xu , Chunxin Zheng , Yulin Li , Zhihai Bi , Jun Ma This is my paper

Pith reviewed 2026-05-19 03:17 UTC · model grok-4.3

classification 💻 cs.RO
keywords wheel-legged robothybrid locomotionreinforcement learning controlrobot joint designsteering performanceterrain adaptationmulti-modal movement
0
0 comments X

The pith

A wheel-legged robot replaces front-leg hip-roll joints with hip-yaw joints to gain efficient flat-ground rolling and adaptive rough-terrain walking.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces FLORES, a wheel-legged robot whose front legs use hip-yaw degrees of freedom in place of the conventional hip-roll joint. This mechanical choice is paired with a reinforcement-learning controller that adapts the Hybrid Internal Model and uses a custom reward structure matched to the new joints. The resulting system produces smooth switches between wheeled and legged motion plus new gaits that combine the strengths of both modes. Experiments show improved steering, lower navigation cost, and reliable performance on mixed surfaces.

Core claim

By substituting hip-yaw actuators for the usual hip-roll degree of freedom on the front legs and training an adapted Hybrid Internal Model controller with rewards tuned to this layout, the robot generates efficient multi-modal locomotion that exploits both wheel rolling on level ground and leg stepping on irregular terrain.

What carries the argument

The front-leg joint reconfiguration that replaces hip-roll with hip-yaw DoFs, together with the customized RL controller built on the Hybrid Internal Model.

Load-bearing premise

The mechanical swap of hip-roll for hip-yaw joints plus the tailored reward function will produce the claimed efficient gaits and steering gains without creating new stability or control problems.

What would settle it

A side-by-side comparison of turning radius and energy cost on flat ground, plus balance recovery time after a sudden slope change, would show whether the new joint layout actually outperforms standard hip-roll designs or introduces hidden instability.

Figures

Figures reproduced from arXiv: 2507.22345 by Chunxin Zheng, Jinglan Xu, Jun Ma, Yulin Li, Zhicheng Song, Zhihai Bi.

Figure 1
Figure 1. Figure 1: FLORES: A reconfigured wheel-legged robot designed for enhanced steering and adaptability across diverse terrains. The structural design of current wheel-legged robots gen￾erally falls into two distinct categories, each with its own set of limitations. The first category treats the legged structure as an active suspension system. This design often results in fewer actuators [8] or limited joint ranges [9],… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of mechanical design: (a) Illustration of the joint DoF [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of our system architecture and training pipeline. The training process consists of two phases: Phase I involves sim-to-sim transfer, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The top figure illustrates Flores climbing up the stairs, while the bottom figure depicts the joint angles of the thigh and calf joints for both the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: By utilizing the modified hip-yaw degree of freedom, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: FLORES is commanded to follow circular paths with a radius of 0.5 m, 1.0 m, 1.5 m, and 2.0 m. The modified hip-yaw DoF, highlighted by the blue line, serves as a steering mechanism that enables FLORES to change its heading angle while keeping the front legs on the ground. This allows the efficient turning without lifting the front legs, thereby reducing overall energy consumption [PITH_FULL_IMAGE:figures/… view at source ↗
Figure 7
Figure 7. Figure 7: FLORES and B2W are performing lateral walking. The blue circles indicate the wheels in contact with the ground, while the red circles represent the wheels that are lifted off the ground. The modified hip-yaw degree of freedom in FLORES allows it to utilize the front wheels to assist in lateral movement. In contrast, B2W must continuously lift its legs to achieve lateral motion, resulting in a less efficien… view at source ↗
Figure 9
Figure 9. Figure 9: FLORES is executing a complex path-following task that necessi￾tates continuous and rapid changes in direction. (a) illustrates the path, with points A, B, C, D, E, F, and G marking the seven points where the robot rapidly changes its direction. (b) and (c) display the instantaneous CoT for FLORES and B2W as they navigate the same path-following task. The spikes in the plots correspond to the turning point… view at source ↗
read the original abstract

Wheel-legged robots integrate leg agility on rough terrain with wheel efficiency on flat ground. However, most existing designs do not fully capitalize on the benefits of both legged and wheeled structures, which limits overall system flexibility and efficiency. We present FLORES, a novel wheel-legged robot design featuring a distinctive front-leg configuration that sets it beyond standard design approaches. Specifically, FLORES replaces the conventional hip-roll degree of freedom (DoF) of the front leg with hip-yaw DoFs, and this allows for efficient movement on flat surfaces while ensuring adaptability when navigating complex terrains. This innovative design facilitates seamless transitions between different locomotion modes (i.e., legged locomotion and wheeled locomotion) and optimizes the performance across varied environments. To fully exploit \flores's mechanical capabilities, we develop a tailored reinforcement learning (RL) controller that adapts the Hybrid Internal Model (HIM) with a customized reward structure optimized for our unique mechanical configuration. This framework enables the generation of adaptive, multi-modal locomotion strategies that facilitate smooth transitions between wheeled and legged movements. Furthermore, our distinctive joint design enables the robot to exhibit novel and highly efficient locomotion gaits that capitalize on the synergistic advantages of both locomotion modes. Through comprehensive experiments, we demonstrate FLORES's enhanced steering capabilities, improved navigation efficiency, and versatile locomotion across various terrains. The open-source project can be found at https://github.com/ZhichengSong6/FLORES.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents FLORES, a wheel-legged robot that replaces the conventional hip-roll DoF in the front legs with hip-yaw DoFs. This reconfiguration is claimed to enable efficient flat-surface movement, terrain adaptability, and seamless transitions between legged and wheeled locomotion modes. A tailored RL controller adapting the Hybrid Internal Model (HIM) with a customized reward structure is introduced to generate novel efficient gaits. Comprehensive experiments are reported to demonstrate enhanced steering capabilities, improved navigation efficiency, and versatile performance across terrains. The project is released as open source.

Significance. If the performance gains can be isolated to the mechanical reconfiguration, the design offers a low-complexity approach to improving steering in hybrid robots while retaining adaptability. The open-source release is a clear strength that supports reproducibility and community follow-up on multi-modal locomotion controllers.

major comments (2)
  1. [Abstract] Abstract: the central claim attributes enhanced steering, novel gaits, and seamless mode transitions to the replacement of hip-roll with hip-yaw DoFs. However, the controller is explicitly a customized RL policy whose reward structure is optimized for this exact mechanical configuration. No ablation (standard hip-roll hardware + identical RL, or new hardware + standard controller) is described that would separate the kinematic effect from reward engineering. This leaves the attribution of gains to the DoF swap unsupported.
  2. [Experiments] Experimental section: the abstract states that comprehensive experiments demonstrate the claimed improvements, yet the manuscript provides neither full methods details, quantitative performance tables with error bars, nor statistical analysis of the reported gains. Without these, the support for the central claims remains preliminary.
minor comments (2)
  1. [Abstract] Clarify the expansion of the acronym FLORES on first use.
  2. [Introduction] Ensure prior wheel-legged robot designs with comparable DoF choices are cited to contextualize the novelty of the front-leg reconfiguration.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful comments, which have helped us improve the clarity and rigor of our manuscript. We address each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim attributes enhanced steering, novel gaits, and seamless mode transitions to the replacement of hip-roll with hip-yaw DoFs. However, the controller is explicitly a customized RL policy whose reward structure is optimized for this exact mechanical configuration. No ablation (standard hip-roll hardware + identical RL, or new hardware + standard controller) is described that would separate the kinematic effect from reward engineering. This leaves the attribution of gains to the DoF swap unsupported.

    Authors: We agree that an explicit ablation study isolating the hardware reconfiguration from the controller would provide stronger evidence for attributing performance gains specifically to the DoF change. Our work presents an integrated system where the mechanical design enables unique capabilities, such as the observed novel gaits, which the tailored RL controller is designed to exploit. Building an additional hardware prototype with standard hip-roll DoFs for comparison was not feasible within the scope of this study. In the revised manuscript, we will revise the abstract and introduction to more precisely state that the gains are demonstrated in the combined hardware-controller system, and add a limitations section acknowledging the lack of hardware ablations as a point for future work. revision: partial

  2. Referee: [Experiments] Experimental section: the abstract states that comprehensive experiments demonstrate the claimed improvements, yet the manuscript provides neither full methods details, quantitative performance tables with error bars, nor statistical analysis of the reported gains. Without these, the support for the central claims remains preliminary.

    Authors: We appreciate this observation and acknowledge that the current experimental presentation could be strengthened with more detailed quantitative data. In the revised version, we will expand the experimental section to include comprehensive methods details, tables summarizing performance metrics (e.g., steering angle, energy efficiency, traversal time) with means and standard deviations from repeated trials, and appropriate statistical analyses to validate the improvements over baseline methods. revision: yes

Circularity Check

0 steps flagged

Mechanical reconfiguration and tailored RL remain distinct; no reduction by construction

full rationale

The paper grounds its claims in an explicit mechanical change (replacing conventional front-leg hip-roll DoF with hip-yaw DoFs) plus a separately trained RL policy that uses a customized reward structure chosen to exploit that hardware. No equations, first-principles derivations, or fitted parameters are shown to be equivalent to the reported performance metrics by construction. Experimental results on physical hardware serve as external validation rather than a closed loop that renames inputs as predictions. No load-bearing self-citations or uniqueness theorems imported from prior author work appear in the derivation chain. This configuration is therefore self-contained against external benchmarks and receives only a minimal circularity score.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work rests on standard assumptions from robotics and RL (actuator torque limits, simulation-to-real transfer, convergence of policy optimization) without introducing new free parameters, axioms, or invented entities in the abstract; the central claim is supported by empirical demonstration rather than derivation from first principles.

pith-pipeline@v0.9.0 · 5799 in / 1126 out tokens · 34673 ms · 2026-05-19T03:17:52.935863+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 1 internal anchor

  1. [1]

    ANYmal-toward legged robots for harsh environments,

    M. Hutter, C. Gehring, A. Lauber, F. Gunther, C. D. Bellicoso, V . Tsounis, P. Fankhauser, R. Diethelm, S. Bachmann, M. Bl ¨osch et al. , “ANYmal-toward legged robots for harsh environments,” Advanced Robotics , vol. 31, no. 17, pp. 918–931, 2017

    work page 2017

  2. [2]

    ://arxiv.org/abs/1909.06586

    D. Kim, J. Di Carlo, B. Katz, G. Bledt, and S. Kim, “Highly dynamic quadruped locomotion via whole-body impulse control and model predictive control,” arXiv preprint arXiv:1909.06586 , 2019

    work page arXiv 1909

  3. [3]

    Keep Rollin’—whole-body motion con- trol and planning for wheeled quadrupedal robots,

    M. Bjelonic, C. D. Bellicoso, Y . de Viragh, D. Sako, F. D. Tresoldi, F. Jenelten, and M. Hutter, “Keep Rollin’—whole-body motion con- trol and planning for wheeled quadrupedal robots,” IEEE Robotics and Automation Letters , vol. 4, no. 2, pp. 2116–2123, 2019

    work page 2019

  4. [4]

    Rolling in the deep–hybrid locomotion for wheeled-legged robots using online trajectory optimization,

    M. Bjelonic, P. K. Sankar, C. D. Bellicoso, H. Vallery, and M. Hutter, “Rolling in the deep–hybrid locomotion for wheeled-legged robots using online trajectory optimization,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 3626–3633, 2020

    work page 2020

  5. [5]

    Flexible motion framework of the six wheel-legged robot: Experimental results,

    S. Wang, Z. Chen, J. Li, J. Wang, J. Li, and J. Zhao, “Flexible motion framework of the six wheel-legged robot: Experimental results,” IEEE/ASME Transactions on Mechatronics , vol. 27, no. 4, pp. 2246– 2257, 2022

    work page 2022

  6. [6]

    Nimbro rescue: Solv- ing disaster-response tasks with the mobile manipulation robot mo- maro,

    M. Schwarz, T. Rodehutskors, D. Droeschel, M. Beul, M. Schreiber, N. Araslanov, I. Ivanov, C. Lenz, J. Razlaw, S. Sch ¨uller, D. Schwarz, A. Topalidou-Kyniazopoulou, and S. Behnke, “Nimbro rescue: Solv- ing disaster-response tasks with the mobile manipulation robot mo- maro,” Journal of Field Robotics , vol. 34, no. 2, p. 400–425, Mar. 2017

    work page 2017

  7. [7]

    Wave: Last mile delivery robotic platform with stair-climbing ability via four-bar linkage-based locomotion,

    S. Park, S. Ahn, J. Shin, H. Kim, J. Yang, Y . Kim, K. Lim, and T. Seo, “Wave: Last mile delivery robotic platform with stair-climbing ability via four-bar linkage-based locomotion,” IEEE/ASME Transactions on Mechatronics, vol. 29, no. 1, pp. 313–323, 2023

    work page 2023

  8. [8]

    Parameters uncertainty analysis of posture control of a four-wheel-legged robot with series slow active suspension system,

    L. Ni, L. Wu, and H. Zhang, “Parameters uncertainty analysis of posture control of a four-wheel-legged robot with series slow active suspension system,” Mechanism and Machine Theory , vol. 175, p. 104966, 2022

    work page 2022

  9. [9]

    Ascento: A two-wheeled jumping robot,

    V . Klemm, A. Morra, C. Salzmann, F. Tschopp, K. Bodie, L. Gulich, N. K ¨ung, D. Mannhart, C. Pfister, M. Vierneisel et al. , “Ascento: A two-wheeled jumping robot,” in 2019 International Conference on Robotics and Automation , 2019, pp. 7515–7521

    work page 2019

  10. [10]

    Actively articulated suspension for a wheel-on-leg rover operating on a martian analog surface,

    W. Reid, F. J. P ´erez-Grau, A. H. G ¨okto˘gan, and S. Sukkarieh, “Actively articulated suspension for a wheel-on-leg rover operating on a martian analog surface,” in 2016 IEEE International Conference on Robotics and Automation , 2016, pp. 5596–5602

    work page 2016

  11. [11]

    CENTAURO: A hybrid locomotion and high power resilient manipu- lation platform,

    N. Kashiri, L. Baccelliere, L. Muratore, A. Laurenzi, Z. Ren, E. M. Hoffman, M. Kamedula, G. F. Rigano, J. Malzahn, S. Cordasco et al., “CENTAURO: A hybrid locomotion and high power resilient manipu- lation platform,” IEEE Robotics and Automation Letters , vol. 4, no. 2, pp. 1595–1602, 2019

    work page 2019

  12. [13]

    Learning-based balance control of wheel-legged robots,

    L. Cui, S. Wang, J. Zhang, D. Zhang, J. Lai, Y . Zheng, Z. Zhang, and Z.-P. Jiang, “Learning-based balance control of wheel-legged robots,” IEEE Robotics and Automation Letters , vol. 6, no. 4, pp. 7667–7674, 2021

    work page 2021

  13. [14]

    Design and research of all-terrain wheel-legged robot,

    J. Zhao, T. Han, S. Wang, C. Liu, J. Fang, and S. Liu, “Design and research of all-terrain wheel-legged robot,” Sensors, vol. 21, no. 16, p. 5367, 2021

    work page 2021

  14. [15]

    Neural networks-based sliding mode tracking control for the four wheel-legged robot under uncertain interaction,

    J. Li, Q. Wu, J. Wang, and J. Li, “Neural networks-based sliding mode tracking control for the four wheel-legged robot under uncertain interaction,” International Journal of Robust and Nonlinear Control , vol. 31, no. 9, pp. 4306–4323, 2021

    work page 2021

  15. [16]

    Whole-body nonlinear model predictive control through contacts for quadrupeds,

    M. Neunert, M. St ¨auble, M. Giftthaler, C. D. Bellicoso, J. Carius, C. Gehring, M. Hutter, and J. Buchli, “Whole-body nonlinear model predictive control through contacts for quadrupeds,” IEEE Robotics and Automation Letters , vol. 3, no. 3, pp. 1458–1465, 2018

    work page 2018

  16. [17]

    Robust rough-terrain locomotion with a quadrupedal robot,

    P. Fankhauser, M. Bjelonic, C. Dario Bellicoso, T. Miki, and M. Hut- ter, “Robust rough-terrain locomotion with a quadrupedal robot,” in 2018 IEEE International Conference on Robotics and Automation , 2018, pp. 5761–5768

    work page 2018

  17. [18]

    Dy- namic locomotion through online nonlinear motion optimization for quadrupedal robots,

    C. D. Bellicoso, F. Jenelten, C. Gehring, and M. Hutter, “Dy- namic locomotion through online nonlinear motion optimization for quadrupedal robots,” IEEE Robotics and Automation Letters , vol. 3, no. 3, pp. 2261–2268, 2018

    work page 2018

  18. [19]

    Mini cheetah: A platform for push- ing the limits of dynamic quadruped control,

    B. Katz, J. D. Carlo, and S. Kim, “Mini cheetah: A platform for push- ing the limits of dynamic quadruped control,” in 2019 International Conference on Robotics and Automation , 2019, pp. 6295–6301

    work page 2019

  19. [20]

    Mit cheetah 3: Design and control of a robust, dynamic quadruped robot,

    G. Bledt, M. J. Powell, B. Katz, J. Di Carlo, P. M. Wensing, and S. Kim, “Mit cheetah 3: Design and control of a robust, dynamic quadruped robot,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2018, pp. 2245–2252

    work page 2018

  20. [21]

    Whleaper: A 10- dof flexible bipedal wheeled robot,

    Y . Zhu, S. He, Z. Qi, Z. Yong, Y . Qin, and J. Chen, “Whleaper: A 10- dof flexible bipedal wheeled robot,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2024, pp. 11 272– 11 277

    work page 2024

  21. [22]

    Reinforcement learning for blind stair climbing with legged and wheeled-legged robots,

    S. Chamorro, V . Klemm, M. d. l. I. Valls, C. Pal, and R. Siegwart, “Reinforcement learning for blind stair climbing with legged and wheeled-legged robots,” arXiv preprint arXiv:2402.06143 , 2024

    work page arXiv 2024

  22. [23]

    Hybrid internal model: Learning agile legged locomotion with simulated robot response,

    J. Long, Z. Wang, Q. Li, L. Cao, J. Gao, and J. Pang, “Hybrid internal model: Learning agile legged locomotion with simulated robot response,” in The 12th International Conference on Learning Representations, 2024

    work page 2024

  23. [24]

    Anytime hybrid driving-stepping loco- motion planning,

    T. Klamt and S. Behnke, “Anytime hybrid driving-stepping loco- motion planning,” in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2017, pp. 4444–4451

    work page 2017

  24. [25]

    Supervised autonomous locomotion and manipulation for disaster response with a centaur-like robot,

    T. Klamt, D. Rodriguez, M. Schwarz, C. Lenz, D. Pavlichenko, D. Droeschel, and S. Behnke, “Supervised autonomous locomotion and manipulation for disaster response with a centaur-like robot,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2018, pp. 1–8

    work page 2018

  25. [26]

    On the kinematics of wheeled motion control of a hybrid wheeled-legged centauro robot,

    M. Kamedula, N. Kashiri, and N. G. Tsagarakis, “On the kinematics of wheeled motion control of a hybrid wheeled-legged centauro robot,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2018, pp. 2426–2433

    work page 2018

  26. [27]

    Robot system of DRC-HUBO+ and control strategy of team KAIST in DARPA robotics challenge finals,

    J. Lim, I. Lee, I. Shim, H. Jung, H. M. Joe, H. Bae, O. Sim, J. Oh, T. Jung, S. Shin et al. , “Robot system of DRC-HUBO+ and control strategy of team KAIST in DARPA robotics challenge finals,”Journal of Field Robotics , vol. 34, no. 4, pp. 802–829, 2017

    work page 2017

  27. [28]

    Whole-body MPC and online gait sequence generation for wheeled-legged robots,

    M. Bjelonic, R. Grandia, O. Harley, C. Galliard, S. Zimmermann, and M. Hutter, “Whole-body MPC and online gait sequence generation for wheeled-legged robots,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2021, pp. 8388–8395

    work page 2021

  28. [29]

    LQR-assisted whole-body control of 9 a wheeled bipedal robot with kinematic loops,

    V . Klemm, A. Morra, L. Gulich, D. Mannhart, D. Rohr, M. Kamel, Y . de Viragh, and R. Siegwart, “LQR-assisted whole-body control of 9 a wheeled bipedal robot with kinematic loops,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 3745–3752, 2020

    work page 2020

  29. [30]

    Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning

    V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa et al. , “Isaac gym: High performance GPU-based physics simulation for robot learning,” arXiv preprint arXiv:2108.10470 , 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  30. [31]

    MuJoCo: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “MuJoCo: A physics engine for model-based control,” in 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2012, pp. 5026–5033

    work page 2012

  31. [32]

    Learning robust autonomous navigation and locomotion for wheeled- legged robots,

    J. Lee, M. Bjelonic, A. Reske, L. Wellhausen, T. Miki, and M. Hutter, “Learning robust autonomous navigation and locomotion for wheeled- legged robots,” Science Robotics , vol. 9, no. 89, p. eadi9641, 2024

    work page 2024