Pith. sign in

REVIEW 3 major objections 4 minor 27 references

RAKOMO: Reachability-Aware K-Order Markov Path Optimization for Quadrupedal Loco-Manipulation

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read RAKOMO integrates a learned reachability margin into K-order Markov Optimization so a reduced-order planner can keep a quadruped's legs within kinematic limits during loco-manipulation, without modeling the legs.

desk verdict A sound incremental integration of learned reachability margins into KOMO; the math is clean, but the evidence is too thin to fully back the headline claim. read the letter →

arxiv 2507.19652 v1 pith:SWBEDSHA submitted 2025-07-25 cs.RO cs.SYeess.SY

Mattia Risiglione , Abdelrahman Abdalla , Victor Barasuol , Kim Tien Ly , Ioannis Havoutis , Claudio Semini This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords trajectoryoptimizationleggedloco-manipulationK-orderMarkovreachabilitymarginquadrupedrobotmotionplanningneuralnetworksurrogatewhole-bodycontrol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Legged manipulators sit between two planning traditions: full whole-body models are accurate but slow, and reduced-order base-only templates are fast but ignore leg joint limits. RAKOMO's claim is that a learned surrogate can close that gap. The paper distills the leg reachable region into a neural network that predicts the reachability margin—the shortest distance from the base's horizontal projection to the region's boundary—and then uses that margin both as a cost and as a safety constraint inside a KOMO optimizer that plans only the base and the arm. In simulated pick-and-place tasks on a HyQReal quadruped with a Kinova Gen3 arm, the margin-aware planner keeps a hip-flexion joint in the middle of its range while the baseline drives it toward its limit, at roughly 2.5 times the baseline compute. The paper's central assertion is that this combination gives gradient-based trajectory optimization the kinematic feasibility of a full model without the computational cost of modeling the legs.

What carries the argument

The load-bearing object is the reachability margin $m(q_t)$, defined as the shortest distance from the horizontal projection of the base to the boundary of the reachable region (the set of base positions, for fixed foot positions and base orientation, where every leg joint respects its limits). Because the exact ray-casting computation is non-differentiable and expensive, the paper trains a three-hidden-layer MLP with 5 million samples drawn around nominal operating values, and feeds the network's gradient into the KOMO cost and constraint Jacobians through Eqs. (13)–(14). The KOMO structure—banded cost/constraint Jacobians and a banded-symmetric Hessian—keeps the added margin terms computationally cheap, so the reduced-order template (base plus arm only) inherits leg kinematic feasibility. The margin thus does the work that an explicit model of the four legs would otherwise do.

What would settle it

Recompute the true reachability margin with the exact ray-casting algorithm at every base pose visited by the RAKOMO-planned trajectories in the two simulations, and check whether the MLP-predicted margin differs from the true value by more than the 0.05 m safety threshold at any of those poses.

Watch

Extended reading notes

Core claim

The core discovery is that the non-differentiable, 30–40 ms ray-casting computation of the leg reachable region can be replaced by an MLP surrogate whose gradient is cheap enough (about 100 ms) to be inserted into a KOMO solver, and that this surrogate, used as the cost $\|\epsilon^* - m(q_t)\|^2$ and the constraint $\epsilon - m(q_t) \le 0$, guides the planner to base poses that keep the legs kinematically feasible. The paper reports that in both a low-object grasp and a high-shelf pick-and-place, the reachability-aware planner holds the relevant hip-flexion-extension joint near the center of its travel, whereas the plain KOMO baseline lowers or raises the trunk so much that the same joint approaches or exceeds its limit. The stated conclusion is that RAKOMO enables rapid convergence of gradient-based motion planning while adapting it effectively to legged manipulators and successfully executing loco-manipulation tasks, with the leg constraints entering only through the learned margin.

Load-bearing premise

The load-bearing premise is that the MLP's predicted reachability margin is an accurate surrogate of the true leg reachable region over the operating domain, and that the gradient used in Eqs. (13)–(14) reliably steers the KOMO solver to base configurations that keep every leg joint within its limits when the whole-body controller executes the plan.

Editorial extensions

If this is right

  • For the two simulated tasks, RAKOMO's planned base poses keep the critical hip joints inside their kinematic limits during execution under the whole-body controller, while baseline KOMO does not.
  • Because the planner uses only the reduced base-plus-arm template, solver time stays at 270–285 ms per solution—about 2.5 times the simplified baseline and far below the estimated ~7x cost of including four leg chains explicitly.
  • The method is compatible with black-box locomotion controllers, since the plan can be executed through velocity-level base commands and arm references without leg-level control.
  • The margin threshold $\epsilon$ gives a tunable safety margin: lowering it makes the planner more aggressive about using the leg workspace, raising it keeps the base more conservative.
  • Retraining the MLP on another quadruped's reachable region transfers the same pipeline to a different platform.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's own compute comparison is against a simple KOMO baseline, not against a full-leg model; a direct empirical benchmark against whole-body KOMO with leg joints would settle how much complexity the surrogate really saves.
  • Because the MLP is trained only around nominal operating values and the paper states that generalization outside that range is not needed, the method's robustness to large pitch, roll, or height excursions during a task remains an open, testable question.
  • The reachable region is recomputed on the fly for each foot configuration; conditioning the margin MLP on foot positions rather than on nominal values would let the same idea extend to walking gaits with moving feet, which the paper lists as future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper presents RAKOMO, which extends K-Order Markov Optimization (KOMO) for quadrupedal loco-manipulation by adding a learned reachability margin. The planner uses a simplified template consisting of the base and the arm, and an MLP trained on samples of foot positions relative to the base predicts the reachability margin; this margin is embedded in the KOMO cost (Eq. 7) and inequality constraint (Eq. 12), with the required gradients supplied through the MLP (Eqs. 13-14). The method is evaluated in two Gazebo simulations with the HyQReal quadruped and a Kinova Gen3 arm: grasping a low object and pick-and-place on a high shelf. Compared with a baseline KOMO planner, RAKOMO keeps the reported reachability margin higher in the first scenario and keeps the right-front HFE joint away from its lower limit in the second, at roughly 2.6x and 2.4x the baseline solve time (104 to 270 ms and 120 to 285 ms). The paper stops short of hardware experiments and does not report repeated trials, quantitative success metrics, or a validation of the learned margin surrogate.

Significance. The core idea is appealing and well motivated: RAKOMO gives a reduced-order planner a differentiable proxy for leg kinematic limits, preserves the banded structure of KOMO, and is compatible with black-box locomotion controllers. The formulation in Section II is coherent, the computational overhead relative to the baseline is modest, and the qualitative trends in the two scenarios are in the expected direction. The main value of the paper is as a planning-formulation contribution. However, the paper does not currently validate the surrogate that carries the safety argument: there is no MLP approximation-error quantification, no repeatability study, no quantitative success metric, and no closed-loop evidence that all leg joints respect limits along the executed trajectory. The paper should be credited for clearly stating the reduced-order assumption and for presenting scenarios that expose the leg-limit issue, but the strength of the claims in the abstract and conclusions exceeds what the present evidence supports.

major comments (3)
  1. [Section II-A, Eq. (12)] The MLP is the only mechanism by which leg kinematic feasibility enters the planner, yet the manuscript reports no approximation error of the learned margin, no gradient-accuracy check, and no coverage analysis relating the training distribution (uniform around nominal operating values) to the configurations actually visited by the optimizer and by the 250 Hz interpolator. Because the safety threshold is epsilon = 0.05 m, an MLP bias or variance of comparable size would invalidate the guarantee implied by Eq. (12). The paper should provide quantitative validation of the surrogate over the relevant operating domain, including the margin values along planned and executed trajectories.
  2. [Section IV, Figs. 6-7] The experimental section consists of two deterministic simulation runs without repeated trials, error bars, or quantitative success metrics. The evidence for the central claim is one margin time series and one joint trace (LF HFE in the low-object scenario and RF HFE in the shelf scenario), but a quadruped has 12 leg joints and the claim is that the method respects leg limitations generally. Because the margin is directly optimized, the margin improvement in Fig. 6 is partly by construction; the independent evidence must therefore come from joint-limit satisfaction, which is currently shown for only one joint per scenario. In the low-object scenario the baseline is described as stopping when the belly touches the ground, which is effectively a failure and should be quantified; the manuscript should report success/failure criteria and per-joint limit-violation statistics for all legs.
  3. [Section III and N=15] Eq. (12) is enforced only at the N=15 waypoints, while execution uses linear interpolation between waypoints and a whole-body controller that can deviate from the reference. With task durations around 14 s, the interpolated or executed path can leave the reachable region between waypoints even when the waypoints satisfy the margin. The paper should report the margin and joint-limit violations along the full executed trajectory (e.g., percentage of time in violation and maximum violation) for all legs and for both methods; without this, the statement in the Conclusions that the motions 'do not push the robot's legs beyond kinematic limits' is not supported.
minor comments (4)
  1. [Eqs. (13)-(14)] The variable g in Eq. (13) is not defined; it should refer to the vector of foot positions relative to the base or to the base orientation variables for which the MLP gradient is taken. Eq. (14) also contains a typo: the derivative should be with respect to xwb rather than xwbi, and the notation 'for all xwb in wxwb' should be made consistent.
  2. [Section I] The phrase 'the the reachability margin' should be corrected, and the expression 'ncxnjxN' is missing multiplication operators and should be typeset properly.
  3. [Fig. 6] The axes appear to contain multiple panels (Pitch, Height, Margin, and possibly HFE angle), but the caption and text describe only the margin panel; the panel layout and legends should be made explicit.
  4. [References] Reference [20] lists the author as 'A. Abdelrahman', which is inconsistent with the author list of this paper and with Ref. [15] ('A. Abdalla'); please align the citation.

Circularity Check

1 steps flagged · score 2.0 of 10

No load-bearing circularity: one margin-improvement plot is tautological, but joint-limit and task-success evidence are independent.

  1. self definitional [Section IV-A, Eq. (7) and Fig. 6]
    "Additionally, we want to optimize the reachability margin m around a desired value ϵ∗ along the robot's trajectory ft = ∥ϵ∗ − m(qt)∥2 (7) ... Results showed that introducing the reachability margin regularization in (7) helps the optimizer to find body poses with better leg kinematic margins along the robot's trajectory."

    The margin m plotted in Fig. 6 is the same quantity that cost (7) explicitly drives toward ϵ∗. A trajectory optimized under (7) will, when evaluated with the same m, show margins near ϵ∗ by construction; this is a restatement of the objective rather than an independent discovery. The circularity is limited to this sanity-check plot: the paper also reports a real joint-angle trace (Fig. 7) and task completion, and the MLP is a surrogate for an independently computed geometric quantity, so the central feasibility claim is not forced by the objective.

full rationale

RAKOMO's central derivation is self-contained: the reachability margin is defined geometrically from the robot's leg joint limits, and the MLP is a learned surrogate of that geometric function. Costs (7) and constraint (12) use the surrogate as intended; no parameter is fitted to the task outcome. The baseline comparison is an external control, and the joint-angle trace in Fig. 7 checks the actual simulated joint against its limits, independent of the cost. Citations to the authors' earlier reachable-region work [15], [20] are background prior art, not load-bearing theorems. The only mild circular element is that Fig. 6 reports the same margin quantity that (7) explicitly optimizes, so the margin improvement is expected by construction; this is a sanity check rather than the load-bearing evidence. Overall score 2.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

No new physical or ontological entities are introduced. The MLP is a computational surrogate for an already-defined reachability margin, not a new entity. The free parameters are hand-tuned values and controller gains, not fitted physical constants. The axioms are standard modeling assumptions for the reduced-order template and the simulation setup.

free parameters (6)
  • desired reachability margin epsilon* = 0.15 m
    Set by hand for the low-height grasping task (Section IV-A); the optimizer is penalized for deviating from this value, so the choice directly shapes the resulting motion.
  • minimum margin threshold epsilon = 0.05 m
    Safety threshold in the inequality constraint (12); hand-chosen.
  • arm/base cost weight ratio = 10
    The positional deviation cost for the arm was weighted 10 times lower than for the base (Section IV-A), a hand-tuned knob that biases the optimizer to move the body rather than the arm.
  • KOMO horizon N = 15
    Fixed number of waypoints for all tests (Section IV); not justified by a convergence study.
  • MLP architecture and training set = 512-256-128 hidden neurons, 5e6 samples
    Chosen empirically; dataset sampling distributions around 'nominal operating values' are not fully specified.
  • whole-body controller gains = not reported
    Used in eqs (15)-(17) but values are not reported; standard tuning parameters that affect tracking performance.
assumptions (7)
  • domain assumption The simplified template model (base and arm only) plus the reachability margin is sufficient for planning legged manipulator tasks.
    This is the key modeling choice in Section II; it assumes the reduced model plus margin captures leg constraints accurately enough for planning.
  • domain assumption All four feet remain fixed on the ground and contacts are not optimized during planning.
    Stated in Section IV: 'In all simulations, we considered the robot in a stance configuration, i.e. with all four feet on the ground and wxbfi fixed.' The entire reachable region computation depends on this.
  • domain assumption The MLP generalizes within the training distribution and the optimizer stays inside that distribution.
    Section II-A: 'generalization outside the selected range is not needed.' This is load-bearing if the optimizer explores outside the training distribution.
  • standard math The KOMO Newton solver with line search converges to a feasible local optimum.
    Used to solve (4); no convergence certificate is provided.
  • domain assumption The whole-body controller at 250 Hz can track the interpolated KOMO trajectories.
    Execution success depends on this, eqs (15)-(18); no tracking error analysis is given.
  • domain assumption The minimum distance from the base's horizontal projection to the edges of the reachable region is a valid scalar measure of leg kinematic feasibility.
    Defined in Section II-A and used in eqs (1), (7), (12); assumes a 2D footprint is sufficient.
  • standard math Computing margins to all polygon edges and taking the minimum yields the true margin even for non-convex regions.
    The paper states the region is non-convex and resorts to iterating over all edges; this assumes the polygon representation is exact.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RAKOMO: Reachability-Aware K-Order Markov Path Optimization for Quadrupedal Loco-Manipulation." pith.science (2026). https://pith.science/paper/SWBEDSHA

@misc{pith2026250719652,
  author       = {Pith},
  title        = {Pith review of: RAKOMO: Reachability-Aware K-Order Markov Path Optimization for Quadrupedal Loco-Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SWBEDSHA}},
  note         = {Machine review of arXiv:2507.19652}
}
read the original abstract

Legged manipulators, such as quadrupeds equipped with robotic arms, require motion planning techniques that account for their complex kinematic constraints in order to perform manipulation tasks both safely and effectively. However, trajectory optimization methods often face challenges due to the hybrid dynamics introduced by contact discontinuities, and tend to neglect leg limitations during planning for computational reasons. In this work, we propose RAKOMO, a path optimization technique that integrates the strengths of K-Order Markov Optimization (KOMO) with a kinematically-aware criterion based on the reachable region defined as reachability margin. We leverage a neural-network to predict the margin and optimize it by incorporating it in the standard KOMO formulation. This approach enables rapid convergence of gradient-based motion planning -- commonly tailored for continuous systems -- while adapting it effectively to legged manipulators, successfully executing loco-manipulation tasks. We benchmark RAKOMO against a baseline KOMO approach through a set of simulations for pick-and-place tasks with the HyQReal quadruped robot equipped with a Kinova Gen3 robotic arm.

Figures

Figures reproduced from arXiv: 2507.19652 by the authors.

Figure 1
Figure 1. Simulation screenshot of IIT’s HyQReal with a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the template model used by KOMO for [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the reachable region (grey area on [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Overview of the control architecture for the planning [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Simulation scenarios: grasping low-height object [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Evolution of the HFE joint of the right front leg during [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 24 canonical work pages

  1. [1]

    Sampling-based algorithms for optimal motion planning,

    S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,” The International Journal of Robotics Research, vol. 30, no. 7, pp. 846–894, 2011. [Online]. Available: https://doi.org/10.1177/0278364911406761

    doi:10.1177/0278364911406761 2011

  2. [2]

    R- lgp: A reachability-guided logic-geometric programming framework for optimal task and motion planning on mobile manipulators,

    K. T. Ly, V . Semenov, M. Risiglione, W. Merkt, and I. Havoutis, “R- lgp: A reachability-guided logic-geometric programming framework for optimal task and motion planning on mobile manipulators,” in2024 IEEE International Conference on Robotics and Automation (ICRA) , 2024, pp. 14 917–14 923

    work page 2024

  3. [3]

    Versatile multicontact planning and control for legged loco-manipulation,

    J.-P. Sleiman, F. Farshidian, and M. Hutter, “Versatile multicontact planning and control for legged loco-manipulation,” Science Robotics, vol. 8, no. 81, p. eadg5014, 2023. [Online]. Available: https: //www.science.org/doi/abs/10.1126/scirobotics.adg5014

    doi:10.1126/scirobotics.adg5014 2023

  4. [4]

    Logic-geometric programming: An optimization-based approach to combined task and motion planning,

    M. Toussaint, “Logic-geometric programming: An optimization-based approach to combined task and motion planning,” in Proceedings of the 24th International Conference on Artificial Intelligence , ser. IJCAI’15. AAAI Press, 2015, p. 19301936

    work page 2015

  5. [5]

    Chomp: gradient optimization techniques for efficient motion planning,

    N. Ratliff, M. Zucker, J. A. Bagnell, and S. Srinivasa, “Chomp: gradient optimization techniques for efficient motion planning,” in Proceedings of the 2009 IEEE International Conference on Robotics and Automation, ser. ICRA’09. IEEE Press, 2009, p. 40304035

    work page 2009

  6. [6]

    Stomp: Stochastic trajectory optimization for motion planning,

    M. Kalakrishnan, S. Chitta, E. Theodorou, P. Pastor, and S. Schaal, “Stomp: Stochastic trajectory optimization for motion planning,” in 2011 IEEE International Conference on Robotics and Automation , 2011, pp. 4569–4574

    work page 2011

  7. [7]

    Motion planning with sequential convex optimization and convex collision checking,

    J. Schulman, Y . Duan, J. Ho, A. X. Lee, I. Awwal, H. Bradlow, J. Pan, S. Patil, K. Goldberg, and P. Abbeel, “Motion planning with sequential convex optimization and convex collision checking,”The International Journal of Robotics Research , vol. 33, pp. 1251 – 1270, 2014. [Online]. Available: https://api.semanticscholar.org/CorpusID:893094

    work page 2014

  8. [8]

    Komo: Newton methods for k-order markov con- strained motion problems,

    M. Toussaint, “Komo: Newton methods for k-order markov con- strained motion problems,” in arXiv preprint arXiv:1407.0414 , 2014

    arXiv 2014

Show all 27 references
  1. [9]

    Navigation in the presence of obstacles for an agile autonomous underwater vehicle,

    M. Xanthidis, N. Karapetyan, H. Damron, S. Rahman, J. Johnson, A. OConnell, J. M. OKane, and I. Rekleitis, “Navigation in the presence of obstacles for an agile autonomous underwater vehicle,” in 2020 IEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 892–899

    2020

  2. [10]

    Differentiable collision avoidance using collision primitives,

    S. Zimmermann, M. Busenhart, S. Huber, R. Poranne, and S. Coros, “Differentiable collision avoidance using collision primitives,” 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 8086–8093, 2022. [Online]. Available: https://api.semanticsch...

    2022

  3. [11]

    Go fetch! - dynamic grasps using boston dynamics spot with external robotic arm,

    S. Zimmermann, R. Poranne, and S. Coros, “Go fetch! - dynamic grasps using boston dynamics spot with external robotic arm,” in 2021 IEEE Int. Conf. on Robotics and Automation , 2021, pp. 4488–4494

    2021

  4. [12]

    Robust rough-terrain locomotion with a quadrupedal robot,

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

    2018

  5. [13]

    Learning Feasibility Constraints for Multi-contact Locomotion of Legged Robots,

    J. Carpentier, R. Budhiraja, and N. Mansard, “Learning Feasibility Constraints for Multi-contact Locomotion of Legged Robots,” in Robotics: Science and Systems , vol. Porceedings of Robotics Science and Systems, Cambridge, MA, United States, July 2017, p. 9p. [Online]. Availab...

    2017

  6. [14]

    Feasible region: An actuation-aware extension of the support region,

    R. Orsolino, M. Focchi, S. Caron, G. Raiola, V . Barasuol, D. G. Cald- well, and C. Semini, “Feasible region: An actuation-aware extension of the support region,” IEEE Transactions on Robotics, vol. 36, no. 4, pp. 1239–1255, 2020

    2020

  7. [15]

    An efficient paradigm for feasibility guarantees in legged locomotion,

    A. Abdalla, M. Focchi, R. Orsolino, and C. Semini, “An efficient paradigm for feasibility guarantees in legged locomotion,” IEEE Transactions on Robotics , vol. 39, no. 5, pp. 3499–3515, 2023

    2023

  8. [16]

    Rapid stability margin estimation for contact-rich loco- motion,

    R. Orsolino, S. Gangapurwala, O. Melon, M. Geisert, I. Havoutis, and M. Fallon, “Rapid stability margin estimation for contact-rich loco- motion,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2021, pp. 8485–8492

    2021

  9. [17]

    A tutorial on newton methods for constrained trajectory optimization and relations to slam, gaussian process smoothing, optimal control, and probabilistic inference,

    M. Toussaint, “A tutorial on newton methods for constrained trajectory optimization and relations to slam, gaussian process smoothing, optimal control, and probabilistic inference,” in Geometric and Numerical Foundations of Movements , 2017. [Online]. Available: https://api.se...

    2017

  10. [18]

    Bitkomo: Combining sampling and optimization for fast convergence in optimal motion planning,

    J. Kamat, J. Ortiz-Haro, M. Toussaint, F. T. Pokorny, and A. Orthey, “Bitkomo: Combining sampling and optimization for fast convergence in optimal motion planning,” in 2022 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS) , 2022, pp. 4492–4497

    2022

  11. [19]

    Quadrupedal footstep planning using learned motion models of a black-box controller,

    I. Taouil, G. Turrisi, D. Schleich, V . Barasuol, C. Semini, and S. Behnke, “Quadrupedal footstep planning using learned motion models of a black-box controller,” 2023 IEEE/RSJ Int. Conference on Intelligent Robots and Systems (IROS) , pp. 800–806, 2023. [Online]. Available: h...

    2023

  12. [20]

    The reachable region: a fast kinematic feasibility criterion for legged locomotion,

    A. Abdelrahman, M. Focchi, R. Orsolino, and C. Semini, “The reachable region: a fast kinematic feasibility criterion for legged locomotion,” in I-RIM Conference, 2020

    2020

  13. [21]

    Newton methods for k-order markov constrained mo- tion problems,

    M. Toussaint, “Newton methods for k-order markov constrained mo- tion problems,” 2014

    2014

  14. [22]

    Fcl: A general purpose library for collision and proximity queries,

    J. Pan, S. Chitta, and D. Manocha, “Fcl: A general purpose library for collision and proximity queries,” in 2012 IEEE International Conference on Robotics and Automation , 2012, pp. 3859–3866

    2012

  15. [23]

    Proximity queries and penetration depth compu- tation on 3d game objects,

    G. Van Den Bergen, “Proximity queries and penetration depth compu- tation on 3d game objects,” in Game developers conference, vol. 170, 2001, p. 209

    2001

  16. [24]

    J. P. Linahan, A Geometric Interpretation of the Boolean Gilbert- Johnson-Keerthi Algorithm. Villanova University, 2015

    2015

  17. [25]

    A whole-body controller based on a simplified template for rendering impedances in quadruped manipulators,

    M. Risiglione, V . Barasuol, D. G. Caldwell, and C. Semini, “A whole-body controller based on a simplified template for rendering impedances in quadruped manipulators,” in 2022 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS), 2022, pp. 9620–9627

    2022

  18. [26]

    Komo tutorial,

    M. Toussaint, “Komo tutorial,” 2023. [Online]. Available: https: //marctoussaint.github.io/robotics-course/

    2023

  19. [27]

    A tutorial on newton methods for constrained trajectory optimization and relations to slam, gaussian process smoothing, optimal control, and probabilistic inference,

    T. Marc, “A tutorial on newton methods for constrained trajectory optimization and relations to slam, gaussian process smoothing, optimal control, and probabilistic inference,” Geometric and Numerical Foundations of Movements , 2017. [Online]. Available: https://doi.org/10.100...

    2017 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.