Pith. sign in

REVIEW 4 major objections 4 minor 28 references

Simultaneous Contact Sequence and Patch Planning for Dynamic Locomotion

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A tree search plus whole-body optimization selects contact sequence, landing patches, and motion in one pass.

desk verdict A genuinely useful MCTS + whole-body TO pipeline for simultaneous contact sequence and patch selection, with real hardware transfer, but the feasibility claims are softer than the abstract suggests. read the letter →

arxiv 2508.12928 v1 pith:OW3YGJAY submitted 2025-08-18 cs.RO

Victor Dhédin , Haizhou Zhao , Majid Khadiv This is my paper · ORCID
classification cs.RO
keywords contactplanningleggedlocomotionMonte-Carlotreesearchwhole-bodytrajectoryoptimizationsequenceselectionpatchacyclicmulti-contactquadrupedandhumanoid
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 locomotion planning usually separates deciding which surfaces to touch and in what order from computing the motion itself, or simplifies the dynamics to make that combined problem tractable. This paper claims that a pipeline combining Monte-Carlo tree search with whole-body trajectory optimization can do both at once: the search proposes contact sequences and landing patches, and the trajectory optimizer checks each candidate against full nonlinear dynamics. For a quadruped crossing a gap or climbing onto a box, the pipeline finds several distinct plans that pass its dynamics and collision checks within seconds, and one box-climbing plan is executed on real hardware under a 30 Hz model-predictive controller. The same pipeline also produces acyclic humanoid maneuvers such as climbing a 0.5 m box. If the claim holds, this closes a gap between discrete contact planning and dynamic motion planning that earlier mixed-integer or gait-fixed methods left open.

What carries the argument

The mechanism is a search-then-verify loop. Monte-Carlo tree search, a tree-search algorithm that balances exploration and exploitation, explores a graph whose nodes are contact modes—for each end-effector, whether it is on the ground and which numbered patch it uses—and whose edges switch feet between keeping, breaking, and making contact, with pruning for minimum contact duration, patch changes without breaking contact, and leg crossing. Each candidate full plan is passed to a whole-body trajectory optimization that uses centroidal momentum dynamics and full kinematics and optimizes states, velocities, and contact forces over N=50 nodes, enforcing non-sliding contact, friction cones, foot-above-surface constraints, patch-boundary limits, and swing-phase collision-avoidance costs. The search reward is the product of a collision term computed from collisions counted at the 50 nodes and a residual term derived from the optimizer's optimality-condition residuals, so approximate dynamic feasibility and collision avoidance are the guiding signals for the discrete search.

What would settle it

Take a plan the pipeline labels collision-free and simulate it in a high-fidelity physics engine with dense time stepping and continuous collision detection, then check whether any link penetrates the environment or the friction cone is violated between the 50 sampled nodes. A single reported plan that fails this check would show the collision and feasibility proxies are permissive; running the same check across the humanoid set would settle how strong the claim is.

Watch

Extended reading notes

Core claim

The central claim is that discrete contact decisions and continuous motion can be planned simultaneously without dropping to linear or centroidal-only dynamics. Each complete contact plan—which end-effectors are in contact at each phase and on which rectangular patch—is scored by running a whole-body trajectory optimization; the optimizer's residuals and the number of collisions counted on the discretized trajectory define the tree-search reward. The paper reports that this identifies many dynamically consistent, collision-free plans for both a gap-crossing task and a box-climbing task across a range of difficulty, with first solutions typically under 10 seconds on easier cases. A planned box climb executed on a real quadruped in a model-predictive control loop demonstrates transfer, and a humanoid variant finds 65 collision-free box-climbing trajectories. The paper states the transferability result as the first demonstration of simultaneous contact sequence and patch selection for acyclic multi-contact locomotion using the whole-body dynamics of a quadruped, while noting that some humanoid trajectories retain non-negligible constraint residuals.

Load-bearing premise

The load-bearing premise is that a plan is accepted as dynamically reliable if the trajectory optimizer reports that its equations are nearly satisfied and if no collision is spotted at the 50 sampled instants; if either check is too lenient, the claimed 'dynamically consistent' and 'collision-free' plans may not actually be feasible.

Editorial extensions

If this is right

  • Because contact modes are generated rather than chosen from a gait library, the same planner can produce acyclic, non-repeating sequences such as using a wall as an intermediate support or climbing onto a box.
  • First solutions arrive within seconds for easier parameter ranges, and several collision-free alternatives per minute are found, so the pipeline can serve as a dataset generator for imitation learning of contact-planning policies.
  • The formulation extends across robot morphologies with minimal changes, so contact-planning search developed on a quadruped transfers to a humanoid with only modified contact models.
  • Because each candidate is verified by whole-body optimization, the final plans are expressed as full state-and-force trajectories ready to be tracked by a model-predictive controller, as demonstrated at 30 Hz replanning on hardware.

Reading between the lines

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

  • The reported per-optimization time and seconds-to-minutes planning times mean the current pipeline is a batch planner, not an online one; a natural extension, which the paper names as future work, is to distill the found plans into a learned policy that selects contact sequences at runtime.
  • Collision checks at 50 discrete nodes leave continuous-time penetration possible between nodes; testing the same reward with swept-volume collision detection or much finer time stepping would turn 'collision-free' into a stronger statement.
  • The reward's reliance on optimizer residuals makes the planner's notion of dynamic feasibility depend on optimizer convergence; as the paper itself notes for some humanoid trajectories, adding a separate post-hoc verification step would harden the pipeline.
  • The same search-then-optimize recipe should transfer to whole-body manipulation, where the end-effectors are hands and patches are graspable surfaces, since the formulation only assumes end-effector contact modes and patch geometry.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The manuscript proposes a pipeline that combines Monte-Carlo Tree Search (MCTS) with whole-body trajectory optimization (TO) to plan contact sequences and contact patches simultaneously for legged robots. The MDP formulation encodes contact modes and patch assignments, graph pruning reduces the search space, and candidate plans are scored by a reward that combines a collision count penalty with a sigmoid of the product of KKT residuals of the trajectory optimization. The TO component uses centroidal momentum dynamics, whole-body kinematics, foot contact constraints, and several end-effector level geometric costs. Experiments cover a Unitree Go2 quadruped crossing gaps up to 1.5 m and climbing boxes up to 0.7 m, with statistics over five runs, one hardware demonstration of a 0.18 m box climb, and a humanoid box-climbing example with 65 reported collision-free trajectories. The paper claims the first demonstration of simultaneous contact sequence and patch selection for acyclic multi-contact locomotion using the whole-body dynamics of a quadruped.

Significance. If the reported validation is tightened, this is a useful contribution to contact planning for legged robots. The pipeline integrates MCTS with whole-body TO in a way that goes beyond earlier centroidal or predefined-gait formulations, and the hardware demonstration, while limited, is a meaningful step beyond pure simulation. The paper also explicitly aims at generating diverse solution sets, which is interesting for downstream imitation-learning applications. However, the central feasibility claims are currently supported by proxies: KKT residuals without a stated acceptance threshold and collision checks at a fixed number of discrete trajectory nodes. The humanoid section itself concedes that some residuals are not negligible. These gaps affect the strongest claims in the abstract and conclusion, so the manuscript needs additional validation before the main claims can be accepted as stated.

major comments (4)
  1. [Section III-A.4, Eq. (3)] The reward R_res uses the product of KKT residuals through a sigmoid, but no acceptance threshold or actual residual values are reported. Section III-A.4 admits that zero constraint violations cannot be guaranteed, and Section IV-D states that some humanoid motions have non-negligible residuals. As a result, the counts of 'collision-free trajectories' in Fig. 6 and the abstract's 'dynamically consistent plans' rest on an unquantified feasibility proxy. Please report the residual distributions, specify the threshold used to accept a plan, and verify a random subset of accepted plans by physics rollout.
  2. [Section III-A.4 and Section IV-B] Collision avoidance is evaluated by counting collisions at the N=50 discretized trajectory nodes in MuJoCo. A fast-moving link can pass through an obstacle between nodes, so this discretization does not establish 'collision-free' in the sense claimed for Fig. 6. Please add continuous collision checking, a denser time grid, or dynamic simulation of the complete plan, and report the resulting success rate over the accepted plans.
  3. [Section IV-C, Fig. 1] The hardware transfer demonstration is a single quadruped climb onto a 0.18 m box, whereas the simulation sweep in Fig. 6 claims plans for box heights up to 0.7 m and gap lengths up to 1.5 m. One easy case does not establish that the pipeline's plans are generally transferable. Please specify the selection procedure for the hardware plan, report additional hardware trials or at least state this as a limitation, and avoid overgeneralizing 'transferable' to the full parameter range.
  4. [Section IV-D] The humanoid results state that 'dynamic feasibility of the trajectories is satisfied for most cases, but not all' and that 'constraint residuals in some of the motions are not negligible.' This directly contradicts a blanket 'dynamically consistent' claim if applied to the humanoid experiments. The abstract and conclusion should be scoped to the accepted quadruped plans, and the residual metric used to decide 'satisfied' should be defined for the humanoid, with the non-negligible cases excluded from the reported counts.
minor comments (4)
  1. [Section IV-B, Figs. 6 and 7] Please clarify whether the five runs use different random seeds and specify how the standard-deviation lines are computed; the axis labels in Figs. 6 and 7 would also benefit from explicit units.
  2. [Section III-B.2] The text states that the reference patch changes halfway through the swing phase, but it does not explain how this switch is encoded in the trajectory optimization, which may be piecewise or require additional decision variables.
  3. [Eq. (11a) and Fig. 4] The relationship between the parameter beta_plus and the resulting swing height is only intuitive; a short derivation or explicit dependence would make the cost easier to interpret.
  4. [Section IV-D] Please clarify the relation between '300 MCTS iterations' and the reported 'first collision-free solution after 52 seconds', since the run time and iteration count are not obviously linked in the text.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the derivation is self-contained; the residual-based reward is a self-consistency proxy, not a fitted prediction or self-citation chain.

full rationale

The paper's central pipeline (MCTS + whole-body TO) is self-contained: contact sequences and patches are searched by MCTS, each candidate is scored by running the TO of Section III-B, and the resulting plans are transferred to hardware. The only potentially circular-looking element is the reward R = R_col R_res, where R_res is a sigmoid of the KKT residuals of the same TO that produces the plan. This is a soft feasibility proxy, not a forced equivalence: the paper explicitly states 'we cannot guarantee zero constraint violations' (Section III-A.4) and reports humanoid cases where 'constraint residuals in some of the motions are not negligible' (Section IV-D), so the residual is an empirically informative measure rather than an identity. The counts of 'collision-free trajectories' in Fig. 6 are based on this proxy, but the central transferability claim is externally validated by the real-robot box-climbing experiment, which uses an MPC built on the same dynamics but executed on physical hardware. Self-citations [20]-[22] are used only to position the contribution (previous MCTS works used simplified dynamics or fixed gaits); they do not supply a load-bearing theorem or uniqueness argument. No parameter is fitted to a target outcome; hyperparameters are hand-tuned. Thus the derivation does not reduce to its inputs by definition or by a self-citation chain.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a standard kino-dynamic model for whole-body planning, a contact-explicit formulation with friction cones, and a residual-based reward that measures feasibility. Hyperparameters include MCTS search parameters, reward scaling, and several cost weights that are hand-tuned. No new physical entities are introduced.

free parameters (8)
  • alpha_col (collision reward scaling) = 0.1
    Scaling in reward R_col (Eq. 2), chosen by hand in Section IV-B.
  • alpha_res (residual reward scaling) = 1/3
    Scaling in reward R_res (Eq. 3), set in Section IV-B.
  • UCB exploration constant C = 1
    Balances exploration/exploitation in MCTS selection (Section IV-B).
  • Minimum contact duration n_phi = 5 (gap), 4 (box)
    Defines phase length in graph pruning (Table I); task-specific.
  • Minimum end-effectors in contact n_min = 0 (gap), 2 (box)
    Prunes transitions with too few contacts (Table I).
  • Cost weights Q (edge, normal, swing) = not reported
    Q_edge, Q_alpha, Q_zsw, Q_xy_sw in Eqs. (8), (10), (11) are chosen by hand but numeric values are not given; the MPC additionally uses retuned weights (Section IV-C).
  • Swing cost parameters beta_plus, beta_minus = 5, 1e-7
    Set the swing height and below-surface penalty in Eq. (11a), given in Fig. 4 caption.
  • Patch edge safety margin epsilon_pi = not reported
    Used in Eq. (8) to keep contacts away from patch edges; value not reported.
assumptions (5)
  • domain assumption The kino-dynamic model (centroidal momentum dynamics with full kinematics, Eq. 4c) is a sufficient model of the robot dynamics for planning executable locomotion.
    Used throughout the TO (Section III-B); if inaccurate, plans may not transfer to hardware.
  • domain assumption The contact-explicit formulation with friction cones and non-sliding constraints (Eqs. 4e-4f) adequately models rigid contact.
    Standard legged robotics assumption; introduced in Section III-B.
  • ad hoc to paper The residual-based reward (Eq. 3) is a reliable proxy for dynamic feasibility of a contact plan.
    Section III-A.4: the authors state zero constraint violations cannot be guaranteed, so plans are ranked by KKT residuals.
  • ad hoc to paper MCTS with graph pruning (minimum contact duration, n_min, no patch change without breaking) does not exclude all feasible or desirable plans for the tested tasks.
    Section III-A.2; if pruning is too aggressive, valid maneuvers are missed.
  • domain assumption MuJoCo collision counting at N=50 discretized trajectory nodes is a sufficient check for collision avoidance.
    Section IV: collisions are counted in MuJoCo during tree search; collisions between nodes could be missed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simultaneous Contact Sequence and Patch Planning for Dynamic Locomotion." pith.science (2026). https://pith.science/paper/OW3YGJAY

@misc{pith2026250812928,
  author       = {Pith},
  title        = {Pith review of: Simultaneous Contact Sequence and Patch Planning for Dynamic Locomotion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OW3YGJAY}},
  note         = {Machine review of arXiv:2508.12928}
}
read the original abstract

Legged robots have the potential to traverse highly constrained environments with agile maneuvers. However, planning such motions requires solving a highly challenging optimization problem with a mixture of continuous and discrete decision variables. In this paper, we present a full pipeline based on Monte-Carlo tree search (MCTS) and whole-body trajectory optimization (TO) to perform simultaneous contact sequence and patch selection on highly challenging environments. Through extensive simulation experiments, we show that our framework can quickly find a diverse set of dynamically consistent plans. We experimentally show that these plans are transferable to a real quadruped robot. We further show that the same framework can find highly complex acyclic humanoid maneuvers. To the best of our knowledge, this is the first demonstration of simultaneous contact sequence and patch selection for acyclic multi-contact locomotion using the whole-body dynamics of a quadruped.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 13 canonical work pages

  1. [1]

    A direct method for trajectory op- timization of rigid bodies through contact,

    M. Posa, C. Cantu, and R. Tedrake, “A direct method for trajectory op- timization of rigid bodies through contact,”The International Journal of Robotics Research, vol. 33, no. 1, pp. 69–81, 2014

    2014

  2. [2]

    Inverse dynamics trajectory optimization for contact-implicit model predictive control,

    V . Kurtz, A. Castro, A. ¨O. ¨Onol, and H. Lin, “Inverse dynamics trajectory optimization for contact-implicit model predictive control,” arXiv preprint arXiv:2309.01813, 2023

    arXiv 2023

  3. [3]

    Fast contact-implicit model predic- tive control,

    S. Le Cleac’h, T. A. Howell, S. Yang, C.-Y . Lee, J. Zhang, A. Bishop, M. Schwager, and Z. Manchester, “Fast contact-implicit model predic- tive control,”IEEE Transactions on Robotics, vol. 40, pp. 1617–1629, 2024

    2024

  4. [4]

    Contact- implicit model predictive control: Controlling diverse quadruped mo- tions without pre-planned contact modes or trajectories,

    G. Kim, D. Kang, J.-H. Kim, S. Hong, and H.-W. Park, “Contact- implicit model predictive control: Controlling diverse quadruped mo- tions without pre-planned contact modes or trajectories,”The Inter- national Journal of Robotics Research, vol. 44, no. 3, pp. 486–510, 2025

    2025

  5. [5]

    Limits of mpcc formula- tions in direct optimal control with nonsmooth differential equations,

    A. Nurkanovi ´c, S. Albrecht, and M. Diehl, “Limits of mpcc formula- tions in direct optimal control with nonsmooth differential equations,” in2020 European control conference (ECC). IEEE, 2020, pp. 2015– 2020

    work page 2020

  6. [6]

    Isaac gym: High performance gpu-based physics simulation for robot learning,

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

    arXiv 2021

  7. [7]

    Mujoco playground,

    K. Zakka, B. Tabanpour, Q. Liao, M. Haiderbhai, S. Holt, J. Y . Luo, A. Allshire, E. Frey, K. Sreenath, L. A. Kahrset al., “Mujoco playground,”arXiv preprint arXiv:2502.08844, 2025

    arXiv 2025

  8. [8]

    Predictive sampling: Real-time behaviour synthesis with mujoco,

    T. Howell, N. Gileadi, S. Tunyasuvunakool, K. Zakka, T. Erez, and Y . Tassa, “Predictive sampling: Real-time behaviour synthesis with mujoco,”arXiv preprint arXiv:2212.00541, 2022

    arXiv 2022

Show all 28 references
  1. [9]

    Real-time whole-body control of legged robots with model- predictive path integral control,

    J. Alvarez-Padilla, J. Z. Zhang, S. Kwok, J. M. Dolan, and Z. Manch- ester, “Real-time whole-body control of legged robots with model- predictive path integral control,”arXiv preprint arXiv:2409.10469, 2024

    2024 arXiv

  2. [10]

    Generative predictive control: Flow matching policies for dynamic and difficult-to-demonstrate tasks,

    V . Kurtz and J. W. Burdick, “Generative predictive control: Flow matching policies for dynamic and difficult-to-demonstrate tasks,” arXiv preprint arXiv:2502.13406, 2025

    2025

  3. [11]

    Learning-based legged locomotion: State of the art and future perspectives,

    S. Ha, J. Lee, M. van de Panne, Z. Xie, W. Yu, and M. Khadiv, “Learning-based legged locomotion: State of the art and future perspectives,”The International Journal of Robotics Research, p. 02783649241312698, 2024

    2024

  4. [12]

    Learning agile loco- motion on risky terrains,

    C. Zhang, N. Rudin, D. Hoeller, and M. Hutter, “Learning agile loco- motion on risky terrains,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 11 864– 11 871

    2024

  5. [13]

    Footstep planning on uneven terrain with mixed-integer convex optimization,

    R. Deits and R. Tedrake, “Footstep planning on uneven terrain with mixed-integer convex optimization,” in2014 IEEE-RAS international conference on humanoid robots. IEEE, 2014, pp. 279–286

    2014

  6. [14]

    Simultaneous contact, gait, and motion planning for robust multilegged locomotion via mixed-integer convex optimization,

    B. Aceituno-Cabezas, C. Mastalli, H. Dai, M. Focchi, A. Radulescu, D. G. Caldwell, J. Cappelletto, J. C. Grieco, G. Fern ´andez-L´opez, and C. Semini, “Simultaneous contact, gait, and motion planning for robust multilegged locomotion via mixed-integer convex optimization,”IEEE...

    2017

  7. [15]

    Efficient multicon- tact pattern generation with sequential convex approximations of the centroidal dynamics,

    B. Ponton, M. Khadiv, A. Meduri, and L. Righetti, “Efficient multicon- tact pattern generation with sequential convex approximations of the centroidal dynamics,”IEEE Transactions on Robotics, vol. 37, no. 5, pp. 1661–1679, 2021

    2021

  8. [16]

    Perceptive mixed-integer footstep control for underactuated bipedal walking on rough terrain,

    B. Acosta and M. Posa, “Perceptive mixed-integer footstep control for underactuated bipedal walking on rough terrain,”arXiv preprint arXiv:2501.19391, 2025

    2025 arXiv

  9. [17]

    Sl1m: Sparse l1-norm minimization for contact plan- ning on uneven terrain,

    S. Tonneau, D. Song, P. Fernbach, N. Mansard, M. Taix, and A. Del Prete, “Sl1m: Sparse l1-norm minimization for contact plan- ning on uneven terrain,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 6604–6610

    2020

  10. [18]

    Perceptive locomotion through whole-body mpc and optimal region selection,

    T. Corb `eres, C. Mastalli, W. Merkt, I. Havoutis, M. Fallon, N. Mansard, T. Flayols, S. Vijayakumar, and S. Tonneau, “Perceptive locomotion through whole-body mpc and optimal region selection,” arXiv preprint arXiv:2305.08926, 2023

    2023 arXiv

  11. [19]

    Monte carlo tree search gait planner for non-gaited legged system control,

    L. Amatucci, J.-H. Kim, J. Hwangbo, and H.-W. Park, “Monte carlo tree search gait planner for non-gaited legged system control,” in2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 4701–4707

    2022

  12. [20]

    Non-gaited legged locomotion with monte-carlo tree search and supervised learning,

    I. Taouil, L. Amatucci, M. Khadiv, A. Dai, V . Barasuol, G. Turrisi, and C. Semini, “Non-gaited legged locomotion with monte-carlo tree search and supervised learning,”IEEE Robotics and Automation Letters, 2024

    2024

  13. [21]

    Diffusion-based learning of contact plans for agile locomotion,

    V . Dh´edin, A. K. C. Ravi, A. Jordana, H. Zhu, A. Meduri, L. Righetti, B. Sch ¨olkopf, and M. Khadiv, “Diffusion-based learning of contact plans for agile locomotion,” in2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids). IEEE, 2024, pp. 637–644

    2024

  14. [22]

    Learning feasible transitions for efficient contact planning,

    R. Akizhanov, V . Dh ´edin, M. Khadiv, and I. Laptev, “Learning feasible transitions for efficient contact planning,”arXiv preprint arXiv:2407.11788, 2024

    2024 arXiv

  15. [23]

    Bandit based monte-carlo planning,

    L. Kocsis and C. Szepesv ´ari, “Bandit based monte-carlo planning,” inEuropean conference on machine learning. Springer, 2006, pp. 282–293

    2006

  16. [24]

    Whole-body motion planning with centroidal dynamics and full kinematics,

    H. Dai, A. Valenzuela, and R. Tedrake, “Whole-body motion planning with centroidal dynamics and full kinematics,” in2014 IEEE-RAS International Conference on Humanoid Robots. IEEE, 2014, pp. 295–302

    2014

  17. [25]

    acados – a modular open-source framework for fast embedded optimal control,

    R. Verschueren, G. Frison, D. Kouzoupis, J. Frey, N. van Duijkeren, A. Zanelli, B. Novoselnik, T. Albin, R. Quirynen, and M. Diehl, “acados – a modular open-source framework for fast embedded optimal control,”Mathematical Programming Computation, 2021

    2021

  18. [26]

    The pinocchio c++ library – a fast and flexible implementation of rigid body dynamics algorithms and their analytical derivatives,

    J. Carpentier, G. Saurel, G. Buondonno, J. Mirabel, F. Lamiraux, O. Stasse, and N. Mansard, “The pinocchio c++ library – a fast and flexible implementation of rigid body dynamics algorithms and their analytical derivatives,” inIEEE International Symposium on System Integration...

    2019

  19. [27]

    CasADi – A software framework for nonlinear optimization and optimal control,

    J. A. E. Andersson, J. Gillis, G. Horn, J. B. Rawlings, and M. Diehl, “CasADi – A software framework for nonlinear optimization and optimal control,”Mathematical Programming Computation, vol. 11, no. 1, pp. 1–36, 2019

    2019

  20. [28]

    Mujoco: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “Mujoco: A physics engine for model-based control,” in2012 IEEE/RSJ international conference on intelligent robots and systems. IEEE, 2012, pp. 5026–5033. Fig. 9. Quaduped robot climbing onto a0.4m high box (top) and crossing a1.1m long gap ...

    2012

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.