pith. sign in

arxiv: 2606.24466 · v2 · pith:XO4G4YIVnew · submitted 2026-06-23 · 💻 cs.RO

FT-WBC: Learning Fault-Tolerant Whole-Body Control for Legged Loco-Manipulation

Yudong Zhong , Pengfei Mai , Sikai Guo , Jiahang Cao , Zhihai Bi , Qiuyue Liu , Ziyan Feng , Jinni Zhou
show 1 more author
Jun Ma
This is my paper

Pith reviewed 2026-06-30 10:13 UTC · model grok-4.3

classification 💻 cs.RO
keywords fault-tolerant controllegged manipulatorswhole-body controlactuator failuresloco-manipulationposture adaptationreinforcement learningrobot stability
0
0 comments X

The pith

FT-WBC predicts actuator faults from lower-body history and adapts base posture to keep legged manipulators stable and functional.

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

Legged manipulators combine walking and arm tasks but become unstable when actuators fail because arm motion shifts the center of mass. The paper presents FT-WBC, which splits control into separate upper- and lower-body policies and adds a fault estimator that reads only leg sensor history to identify failed joints. A posture adaptation module then revises the base posture commands coming from the arm policy so the lower body can still execute them without falling. This produces compensatory gaits that hold the robot upright while keeping the arm workspace as large as possible. Simulation trials and real-robot tests show higher survival rates and retained reach under both weakening and locked failures, with the policy transferring directly from simulation to hardware.

Core claim

FT-WBC adopts a decoupled upper- and lower-body policy architecture and introduces a Fault Estimator that predicts faulty joints from lower-body proprioceptive histories together with a Posture Adaptation Module that converts potentially unstable posture requests into safe base commands, allowing synthesis of compensatory gaits that preserve arm workspace while maintaining whole-body stability under actuator failures.

What carries the argument

The fault-aware posture adaptation mechanism formed by the Fault Estimator (FE) and Posture Adaptation Module (PAM), which modifies base posture plans using predicted faults to produce executable lower-body commands.

If this is right

  • Survival rate under actuator weakening or locking failures increases substantially compared with non-adaptive controllers.
  • Arm workspace is retained to the greatest extent possible while whole-body stability is preserved.
  • The learned policy transfers zero-shot from simulation to a physical legged manipulator.
  • Compensatory gaits are generated on the fly without requiring changes to the lower-body locomotion policy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The separation of fault estimation from the locomotion policy could allow the same estimator to be reused across different lower-body controllers.
  • If the estimator works from leg data alone, similar modules might be added to other robots that combine mobility with manipulation without adding arm sensors.
  • The approach opens the possibility of trading hardware redundancy for software adaptation in safety-critical deployments.

Load-bearing premise

The fault estimator can reliably identify which joints have failed using only lower-body proprioceptive history so that the posture module can generate commands the lower-body policy can execute while staying stable.

What would settle it

A controlled experiment in which the fault estimator is deliberately given incorrect failure labels and the robot is observed to fall or lose workspace at rates no better than a non-adaptive baseline controller.

Figures

Figures reproduced from arXiv: 2606.24466 by Jiahang Cao, Jinni Zhou, Jun Ma, Pengfei Mai, Qiuyue Liu, Sikai Guo, Yudong Zhong, Zhihai Bi, Ziyan Feng.

Figure 1
Figure 1. Figure 1: We present FT-WBC, a framework that enables robust, fault-tolerant whole-body control [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of FT-WBC. The framework adopts a decoupled architecture consisting of a [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Evaluation of the FE. The x-axis denotes time, and the y-axis denotes the joint name. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Forward linear velocity (vx) tracking profiles under severe joint weakening (kτ = 0.1). The target command (dashed grey) is set to 0.4 m/s. The solid grey, red, orange, and blue lines represent Robo-Duet, w/o FE, w/o PAM, and our full framework, respectively. Our method signifi￾cantly suppresses velocity oscillations and maintains tight tracking across diverse joint failures. Velocity tracking [PITH_FULL_… view at source ↗
Figure 5
Figure 5. Figure 5: Real-world dynamic manipulation under different joint failures. Each subfigure shows [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sequential simulation snapshots at T = 0 s and T = 1 s under representative PW, CW, and LF joint failure configurations for velocity tracking (top) and workspace test (bottom). Red crosses indicate the joints with injected failures across the quadruped platform. As shown in [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
read the original abstract

Legged manipulators combine the mobility of legged platforms with the manipulation capability of robotic arms. However, arm-induced Center-of-Mass shifts and dynamic disturbances make the system more prone to instability under actuator failures, potentially leading to falls, task failures, or safety risks. Existing fault-tolerant control methods mainly focus on locomotion alone, leaving the coupled problem of whole-body stability and arm reachability in fault-tolerant loco-manipulation largely unaddressed. To bridge this gap, we propose FT-WBC, a fault-tolerant loco-manipulation framework for robust whole-body control of legged manipulators under actuator failures. FT-WBC adopts a decoupled upper- and lower-body policy architecture and introduces two key modules: a Fault Estimator (FE) and a Posture Adaptation Module (PAM). The FE predicts faulty joints from lower-body proprioceptive histories, while the PAM uses this fault information to adapt the base posture plan generated by the arm policy, converting potentially unstable posture requests into safe and executable base posture commands. Through this fault-aware posture adaptation mechanism, FT-WBC synthesizes compensatory gaits under actuator failures and preserves as much arm workspace as possible while maintaining whole-body stability. Simulation and real-world experiments show that FT-WBC significantly improves survival rate and workspace under weakening or locked failures, and transfers zero-shot to a real legged manipulator in the real world.

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 / 1 minor

Summary. The paper proposes FT-WBC, a learning-based fault-tolerant whole-body control framework for legged manipulators. It uses a decoupled upper- and lower-body policy architecture together with a Fault Estimator (FE) that predicts faulty joints from lower-body proprioceptive histories and a Posture Adaptation Module (PAM) that converts arm-policy posture requests into safe base commands. The central claim is that this fault-aware adaptation enables compensatory gaits under actuator failures while preserving arm workspace and whole-body stability, supported by simulation and real-world experiments showing improved survival rate and workspace.

Significance. If the empirical claims hold and the FE/PAM mechanism is shown to function as described, the work addresses a practically important gap between existing fault-tolerant locomotion methods and coupled loco-manipulation under failure. The decoupled architecture and zero-shot real-world transfer would be notable strengths for the field.

major comments (2)
  1. [Abstract] Abstract: the statement that 'simulation and real-world experiments show that FT-WBC significantly improves survival rate and workspace' supplies no quantitative values, baselines, failure-mode definitions, trial counts, or statistical details. Because the central claim rests on these uninspectable results, the magnitude and reliability of the improvement cannot be assessed.
  2. [Method (Fault Estimator)] Method section describing the Fault Estimator: the FE is asserted to predict faulty joints (including arm actuators) solely from lower-body proprioceptive history. Arm actuator failures produce no direct change in lower-body joint positions, velocities, or torques beyond secondary effects already handled by the locomotion policy. No explicit cross-body correlation mechanism, arm-state reconstruction, or learned mapping is described that would allow reliable detection of arm faults; without such a mechanism the subsequent PAM adaptation for arm failures lacks a sound basis and directly affects the 'preserves as much arm workspace' claim.
minor comments (1)
  1. [Abstract] The abstract is unusually long and contains the main empirical claims; moving quantitative results and failure definitions to a dedicated experiments paragraph would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our results and the technical details of the Fault Estimator. We address each major comment below and will incorporate revisions to improve transparency and completeness.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the statement that 'simulation and real-world experiments show that FT-WBC significantly improves survival rate and workspace' supplies no quantitative values, baselines, failure-mode definitions, trial counts, or statistical details. Because the central claim rests on these uninspectable results, the magnitude and reliability of the improvement cannot be assessed.

    Authors: We agree that the abstract would be strengthened by including quantitative details. In the revised manuscript we will expand the abstract to report specific metrics (e.g., survival-rate improvements under weakening and locked failures, workspace preservation percentages, number of trials, baseline comparisons, and failure-mode definitions) drawn from the simulation and real-world sections. This change will make the central empirical claims directly assessable without altering the reported results. revision: yes

  2. Referee: [Method (Fault Estimator)] Method section describing the Fault Estimator: the FE is asserted to predict faulty joints (including arm actuators) solely from lower-body proprioceptive history. Arm actuator failures produce no direct change in lower-body joint positions, velocities, or torques beyond secondary effects already handled by the locomotion policy. No explicit cross-body correlation mechanism, arm-state reconstruction, or learned mapping is described that would allow reliable detection of arm faults; without such a mechanism the subsequent PAM adaptation for arm failures lacks a sound basis and directly affects the 'preserves as much arm workspace' claim.

    Authors: The Fault Estimator is a learned neural network trained on proprioceptive histories collected under both lower-body and arm actuator failures. Although arm failures lack direct lower-body signatures, they produce observable secondary effects on base dynamics, CoM shifts, and posture stability that are captured in the lower-body state history; the network learns the resulting cross-body mapping during training. We will revise the method section to explicitly describe the training data composition, network architecture, loss function, and how these indirect correlations are learned, thereby providing the mechanistic basis for arm-fault detection and the subsequent PAM adaptation. revision: yes

Circularity Check

0 steps flagged

No circularity; learned architecture with no self-referential derivation

full rationale

The paper describes a decoupled policy architecture (Fault Estimator predicting from lower-body histories, Posture Adaptation Module adapting base posture) trained via reinforcement learning. No equations, fitted parameters, or derivation chain are presented that reduce any claimed output to the inputs by construction. Claims rest on empirical survival rates and workspace metrics in simulation and real-world transfer, which are externally falsifiable and independent of any self-definition or self-citation load-bearing step. The method is architectural rather than a mathematical derivation, so no load-bearing step reduces to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly relies on standard RL training assumptions and the unstated premise that proprioceptive history suffices for fault detection.

pith-pipeline@v0.9.1-grok · 5807 in / 1140 out tokens · 30010 ms · 2026-06-30T10:13:20.881341+00:00 · methodology

Review history (2 revisions) →

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 · 13 canonical work pages

  1. [1]

    A. E. Vijayan, A. Cramariuc, M. Risiglione, C. Gehring, and M. Hutter. Multi-critic learning for whole-body end-effector twist tracking.arXiv preprint arXiv:2507.08656, 2025

    work page arXiv 2025

  2. [2]

    X. Liu, B. Ma, C. Qi, Y . Ding, N. Xu, G. Zhang, P. Chen, K. Liu, Z. Jia, C. Guan, et al. Mlm: Learning multi-task loco-manipulation whole-body control for quadruped robot with arm.IEEE Robotics and Automation Letters, 11(1):81–88, 2025

    2025

  3. [3]

    H. Ha, Y . Gao, Z. Fu, J. Tan, and S. Song. Umi on legs: Making manipulation policies mobile with manipulation-centric whole-body controllers.arXiv preprint arXiv:2407.10353, 2024

    work page arXiv 2024

  4. [4]

    Y . Fu, Y . Zhang, Q. Yang, L. Yan, Z. Cao, and Y . Gao. Contrastive forward prediction rein- forcement learning for adaptive fault-tolerant legged robots. InConference on Robot Learning, pages 3285–3303. PMLR, 2025

    2025

  5. [5]

    M. Liu, D. Pathak, and A. Agarwal. Locoformer: Generalist locomotion via long-context adaptation.arXiv preprint arXiv:2509.23745, 2025

    work page arXiv 2025

  6. [6]

    S. Lee, I. M. A. Nahrendra, D. Lee, B. Yu, M. Oh, H. Lee, and H. Myung. Dreamflex: Learn- ing fault-aware quadrupedal locomotion controller for anomaly situation in rough terrains. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 16001– 16007. IEEE, 2025

    2025

  7. [7]

    Z. Fu, X. Cheng, and D. Pathak. Deep whole-body control: learning a unified policy for manipulation and locomotion. InConference on Robot Learning, pages 138–149. PMLR, 2023

    2023

  8. [8]

    G. Pan, Q. Ben, Z. Yuan, G. Jiang, Y . Ji, S. Li, J. Pang, H. Liu, and H. Xu. Roboduet: Learning a cooperative policy for whole-body legged loco-manipulation.IEEE Robotics and Automation Letters, 2025

    2025

  9. [9]

    N. Fey, G. B. Margolis, M. Peticco, and P. Agrawal. Bridging the sim-to-real gap for athletic loco-manipulation.arXiv preprint arXiv:2502.10894, 2025

    work page arXiv 2025

  10. [10]

    Portela, G

    T. Portela, G. B. Margolis, Y . Ji, and P. Agrawal. Learning force control for legged manip- ulation. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 15366–15372. IEEE, 2024

    2024

  11. [11]

    Q. Li, Q. Meng, Y . Qin, J. Chen, X. Ding, and K. Xu. Dynamic interaction control in legged mobile manipulators: A decoupled approach. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 1406–1412. IEEE, 2024

    2024

  12. [12]

    A. Rigo, M. Hu, S. K. Gupta, and Q. Nguyen. Hierarchical optimization-based control for whole-body loco-manipulation of heavy objects. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 15322–15328. IEEE, 2024

    2024

  13. [13]

    X. Zhu, Y . Chen, L. Sun, F. Niroui, S. L. Cleac’h, J. Wang, and K. Fang. Versatile loco- manipulation through flexible interlimb coordination.arXiv preprint arXiv:2506.07876, 2025

    work page arXiv 2025

  14. [14]

    Y . Niu, Y . Zhang, M. Yu, C. Lin, C. Li, Y . Wang, Y . Yang, W. Yu, T. Zhang, Z. Li, et al. Human2locoman: Learning versatile quadrupedal manipulation with human pretraining.arXiv preprint arXiv:2506.16475, 2025. 9

    work page arXiv 2025

  15. [15]

    Huang, A

    H. Huang, A. Loquercio, A. Kumar, N. Thakkar, K. Goldberg, and J. Malik. Manipulator as a tail: Promoting dynamic stability for legged locomotion. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 9712–9719. IEEE, 2024

    2024

  16. [16]

    M. Liu, Z. Chen, X. Cheng, Y . Ji, R.-Z. Qiu, R. Yang, and X. Wang. Visual whole-body control for legged loco-manipulation.arXiv preprint arXiv:2403.16967, 2024

    work page arXiv 2024

  17. [17]

    Zhang, N

    J. Zhang, N. Gireesh, J. Wang, X. Fang, C. Xu, W. Chen, L. Dai, and H. Wang. Gamma: Graspability-aware mobile manipulation policy learning based on online grasping pose fusion. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 1399–

  18. [18]

    Bimanual robot-assisted dressing: A spherical coordinate-based strategy for tight-fitting garments

    A. Jaafar, S. S. Raman, S. Harithas, Y . Wei, S. Juliani, A. Wernerfelt, B. Quartey, I. Idrees, J. X. Liu, and S. Tellex.λ: A benchmark for data-efficiency in long-horizon indoor mobile manipulation robotics. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 8942–8949, 2025. doi:10.1109/IROS60139.2025.11246789

    work page doi:10.1109/iros60139.2025.11246789 2025

  19. [19]

    Mendonca, E

    R. Mendonca, E. Panov, B. Bucher, J. Wang, and D. Pathak. Continuously improving mobile manipulation with autonomous real-world rl.arXiv preprint arXiv:2409.20568, 2024

    work page arXiv 2024

  20. [20]

    S. Xie, C. Hu, D. Wang, J. Johnson, M. Bagavathiannan, and D. Song. Coupled active percep- tion and manipulation planning for a mobile manipulator in precision agriculture applications. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 12665– 12671. IEEE, 2024

    2024

  21. [21]

    K. Liu, Z. Wang, B. Li, L. Zhu, and H. Ding. Fault joint detection and adaptive fault-tolerant control of legged robots under joint partial failures.IEEE Robotics and Automation Letters, 2025

    2025

  22. [22]

    Z. Luo, E. Xiao, and P. Lu. Ft-net: Learning failure recovery and fault-tolerant locomotion for quadruped robots.IEEE Robotics and Automation Letters, 8(12):8414–8421, 2023

    2023

  23. [23]

    F. Yang, C. Yang, D. Guo, H. Liu, and F. Sun. Fault-aware robust control via adversarial reinforcement learning. In2021 IEEE 11th Annual International Conference on CYBER Tech- nology in Automation, Control, and Intelligent Systems (CYBER), pages 109–115. IEEE, 2021

    2021

  24. [24]

    D. Liu, T. Zhang, J. Yin, and S. See. Saving the limping: Fault-tolerant quadruped locomotion via reinforcement learning.arXiv preprint arXiv:2210.00474, 2022

    work page arXiv 2022

  25. [25]

    Okamoto, H

    W. Okamoto, H. Kera, and K. Kawamoto. Reinforcement learning with adaptive curriculum dynamics randomization for fault-tolerant robot control.arXiv preprint arXiv:2111.10005, 2021

    work page arXiv 2021

  26. [26]

    X. Wu, W. Dong, H. Lai, Y . Yu, and Y . Wen. Adaptive control strategy for quadruped robots in actuator degradation scenarios. InProceedings of the Fifth International Conference on Distributed Artificial Intelligence, pages 1–13, 2023

    2023

  27. [27]

    Zhang, Z

    P. Zhang, Z. Hua, Q. Qiu, and J. Ding. Ft-cpg: Learning central pattern generators for fault- tolerant quadruped locomotion under multi-joint failures.IEEE Robotics and Automation Let- ters, 2025

    2025

  28. [28]

    M. Kim, U. Shin, and J.-Y . Kim. Learning quadrupedal locomotion with impaired joints using random joint masking. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 9751–9757. IEEE, 2024

    2024

  29. [29]

    Y . Qiu, X. Lin, J. Wang, X. Li, L. Qi, and M.-H. Yang. Umc: Unified resilient controller for legged robots with joint malfunctions.arXiv preprint arXiv:2502.03035, 2025. 10

    work page arXiv 2025

  30. [30]

    C. Chen, C. Li, H. Lu, Y . Wang, and R. Xiong. Meta reinforcement learning of locomotion policy for quadruped robots with motor stuck.IEEE Transactions on Automation Science and Engineering, 22:5551–5565, 2024

    2024

  31. [31]

    T. Hou, J. Tu, X. Gao, Z. Dong, P. Zhai, and L. Zhang. Multi-task learning of active fault- tolerant controller for leg failures in quadruped robots. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 9758–9764. IEEE, 2024. 11 Appendix A Leg Policy and Arm Policy A.1 Policy Inputs and Actions FT-WBC adopts a decoupled upper- and...

    work page arXiv 2024