Pith. sign in

REVIEW 4 major objections 5 minor 55 references

Perceptive Mixed-Integer Footstep Control for Underactuated Bipedal Walking on Rough Terrain

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read A single mixed-integer quadratic program can pick footholds, foot placement, ankle torque, and step timing at over 100 Hz, letting an underactuated biped walk on discontinuous terrain

desk verdict A strong systems paper whose abstract overclaims the safety guarantee; the MIQP and S3 are real contributions, but the safety thresholds are hand-tuned and the hardware demo is a single trial. read the letter →

arxiv 2501.19391 v2 pith:Q2YKZ2HT submitted 2025-01-31 cs.RO

classification cs.RO
keywords mixed-integerquadraticprogrammodelpredictivecontrolfootstepplanningbipedalwalkingunderactuatedrobotterrainsegmentationelevationmappingALIP
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

This paper claims that a bipedal robot can choose where to step and how to step at the same time, fast enough for real-time walking on rough ground. The key move is to cast the whole problem—which safe patch of ground to use, where exactly to place the foot, how much ankle torque to apply, and how long to stand—as a single mixed-integer quadratic program over a reduced-order dynamic model, and to show that this program solves in under 10 milliseconds. The paper pairs that controller with a terrain-segmentation method that turns a noisy elevation map into stable convex foothold regions, avoiding the flicker that earlier plane-fitting approaches suffered. If the claim holds, mixed-integer footstep selection with dynamics is not just an offline planning tool but a practical real-time control architecture for underactuated walking on discontinuous, previously unseen terrain.

What carries the argument

The load-bearing object is the model-predictive footstep control (MPFC) problem, a mixed-integer quadratic program: one binary variable per foothold candidate per step chooses which convex polygon the footstep center must lie in, enforced by relaxing each polygon's linear constraints by a large constant $M$ unless that binary variable is 1. The continuous variables are step-to-step states of the Angular Momentum Linear Inverted Pendulum (ALIP)—a reduced-order model of the horizontal center-of-mass dynamics using angular momentum about the contact point—footstep positions $p_n$, an initial ankle torque $u$, and the remaining stance time $T$. The dynamics are the step-to-step ALIP map $x_{n+1}=A_{s2s}x_n+B_{s2s}(p_{n+1}-p_n)$, with the initial step produced from the current state by a linearized timing update; the cost projects the ALIP trajectory onto the subspace of two-step-periodic orbits at the desired walking velocity. The companion machinery is Stable Steppability Segmentation (S3): per-pixel curvature and inclination safety scores, a hysteresis term that carries the previous safe/unsafe classification forward, then approximate convex decomposition with a greedy inner-approximation ('whittling') and a least-squares plane fit, converting an elevation map into the convex foothold set that MPFC consumes.

What would settle it

Run the stack, exactly as in the hardware experiments, on a course with a 16 cm step whose lower edge is filled with dried leaves; if S3 labels the edge as steppable and MPFC commands a foothold there, causing the robot to slip or trip, the central claim that the controller avoids unsafe areas is refuted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that simultaneous optimization over discrete foothold choice and continuous walking dynamics is real-time feasible and hardware-viable. MPFC is a single mixed-integer quadratic program (MIQP): binary variables $\mu_{n,i}$ assign the center of each planned footstep to exactly one convex polygon foothold, while continuous variables—the step-to-step Angular Momentum Linear Inverted Pendulum (ALIP) states, footstep positions, initial ankle torque, and remaining stance time—are coupled through the step-to-step ALIP dynamics and a linearized stance-timing map. The controller runs at over 100 Hz, with a median solve time of 2 ms across 134,654 solves from hardware trials and a worst observed solve of 12.6 ms, and the paper reports the underactuated robot Cassie walking up and down brick steps, over a curb, and up a grassy slope in one continuous outdoor trial. The companion claim is that terrain segmentation can be made temporally consistent by classifying elevation-map pixels safe or unsafe with local curvature and inclination criteria plus hysteresis, and then convexifying the mask, rather than by fitting global planes; this removes the flicker that made earlier plane-segmentation-based footstep planners brittle. Together these establish that perceptive, dynamic, underactuated walking over constrained footholds can be closed with model-based optimization instead of a fixed footstep sequence and offline terrain knowledge.

Load-bearing premise

The load-bearing premise is that the hand-picked safety thresholds and the amount of temporal smoothing in the terrain classifier keep unsafe ground out of the foothold set; the paper's own failure case (Section IX.B.2) shows dried leaves under a step edge being classified as steppable, and one such misclassification can make the planner place a foot on unsafe terrain.

Editorial extensions

If this is right

  • Mixed-integer footstep selection with reduced-order dynamics is a real-time control primitive: median solve time is 2 ms and 99.9% of solves finish in under 7.7 ms, with a worst observed solve of 12.6 ms.
  • A biped can traverse previously unseen discontinuous terrain without a preselected foothold sequence, demonstrated by Cassie walking over brick steps, a curb, and grass in one continuous trial.
  • Decoupling safe-terrain classification from plane fitting removes the main source of temporal inconsistency; frame-to-frame intersection-over-union stays near 1 where plane-segmentation baselines flicker across the whole [0,1] range.
  • Optimizing the initial stance duration improves success on sparse footholds, with the largest gains on the smallest stepping stones.
  • Elevation mapping, S3 segmentation, and convex decomposition run in one CPU thread within the camera-frame budget, so the perception side of the stack is real-time as well as the controller.

Reading between the lines

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

  • Because foothold choice and dynamics share one optimization, the same template could absorb costs that penalize stepping near edges or reward safety margin directly, rather than relying solely on the terrain mask.
  • The paper's lateral reset-map patch treats weight transfer as instantaneous in the coronal plane; measuring lateral angular momentum through double stance on hardware would show whether the model or the perception stack is the binding constraint on step width.
  • S3's safety criteria are plug-in functions, so a learned or semantic steppability classifier could replace the heuristic curvature and inclination scores and likely fix the dried-leaves failure case while keeping the hysteresis machinery.
  • With faster solvers or stronger integer-cut formulations, the same controller could extend beyond the two-step horizon and address the overly optimistic foothold choices the paper lists as an algorithmic limitation.
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

4 major / 5 minor

Summary. The paper proposes a full perception-and-control stack for underactuated bipedal walking on rough terrain. The controller, MPFC, is a mixed-integer quadratic program that jointly optimizes over discrete foothold assignment, continuous footstep positions, ALIP reduced-order dynamics, initial ankle torque, and step timing, and is reported to solve at rates around 100 Hz. The perception stack, S3, converts an elevation map into a hysteresis-based steppability mask and then into convex polygon footholds, which become hard constraints in the MIQP. The claims are supported by simulation sweeps over stepping-stone terrains, solve-time statistics from more than 134,000 solves, offline perception benchmarks against plane-segmentation baselines, and an outdoor hardware demonstration on Cassie over brick steps, a curb, and a grassy slope.

Significance. If the claims hold, this is a meaningful advance: it is among the first demonstrations that a single MIQP over foothold choice, footstep position, template dynamics, and timing can be solved in real time and deployed on hardware with perception in the loop. The paper has genuine strengths: the algebraic derivations in Section III are internally consistent, the reset-map and Bds computations check out, the solve-time dataset is large and clearly reported, the S3 temporal-consistency benchmark is a useful comparison, and the authors state that source code will be released in dairlib. The main gap is not in the control derivation but in the empirical and safety evidence for the perception layer and in a few overreaching performance claims.

major comments (4)
  1. [VI.A, Table VIII, IX.B.2] The abstract's claim that the robot walks 'without stepping in unsafe areas' is not established, because the soundness of the S3 foothold mask rests on two heuristic criteria (Eq. 25 and Eq. 26) and two manually chosen constants, k_safe=0.7 and k_hyst=0.6 (Table VIII), with no reported calibration against ground-truth steppability, Cassie's foot geometry, tracking error, or reachable workspace. The paper's own Section IX.B.2 documents a false positive in which dried leaves under a step edge were classified as steppable; since the big-M constraints in Eq. (20) turn any pixel that survives S3 into a feasible footstep, such a false positive directly invalidates the safety claim. The authors should either qualify the safety claim to the tested conditions or add a validation/calibration procedure for the thresholds, for example derived from foot dimensions and swing tracking bounds.
  2. [III.B, Appendix A] The reset-map derivation depends on the assumption [(pCoP - p-) x m vCoM]_{x,y} ≈ 0, which is justified only when the CoM velocity is nearly parallel to the step vector; for lateral stepping, and for walking on terrain with a lateral component, this condition need not hold and no error bound or sensitivity analysis is given. The hardware implementation further patches the lateral direction in Appendix A by using f(t)=1 and blending Bds terms as a feed-forward correction for hip-roll compliance, which is a heuristic compensation rather than a consequence of the modeling assumptions. Because lateral stabilization is central to the hardware demonstration, the model error of the double-stance reset map for lateral steps should be quantified, either by a simulation study or a dedicated hardware experiment.
  3. [VIII.C, Table II, Abstract] The abstract's 'at over 100 Hz' and the introduction's 'solve times of less than 10 milliseconds' are not fully supported by the reported data: Table II gives a maximum solve time of 12.6 ms over 134,654 solves, even though 99.9% of solves are below 7.7 ms. A hard 100 Hz control loop must accommodate its worst case, so the authors should state explicitly whether the 100 Hz claim refers to the median, the 99.9th percentile, or another statistic, and discuss whether a 12.6 ms solve affects the real-time behavior of the stack.
  4. [VIII.B, Fig. 13] The hardware evidence for the headline result is presented as a single continuous trial ('A single trial traversing steps, a curb, and a grass hill is shown'), with additional footage only in the supplemental video. There are no repeated-trial statistics, failure counts, or quantitative success criteria across terrain variations, which weakens the empirical basis for the 'state of the art' claim. Reporting the number of runs, the number of falls or slips, and the range of terrain parameters would make the hardware claim reproducible and comparable to prior work.
minor comments (5)
  1. [VIII.A, Fig. 12] The stepping-stone success rates are computed over 50 random terrains per condition, but Fig. 12 displays only pointwise success rates without confidence intervals or error bars; adding binomial confidence intervals or shaded bands would materially strengthen the step-timing optimization comparison.
  2. [IV.C] The real-time MIQP results do not report the Gurobi optimality-gap tolerance, node limit, or whether any solves terminate early due to a time limit; since the controller relies on constraint satisfaction rather than global optimality, stating the solver settings would clarify what 'solve time' means.
  3. [IX.A.3] There is a typo in the heading 'Foostep Height Lookup'; it should read 'Footstep Height Lookup'. The same typo appears in the text describing the footstep height adjustment.
  4. [VIII.D.4, Table IV] Table IV shows that with the hardware value k_hyst=0.6, none of the tossed moving obstacles produce a hole in the segmentation before coming to rest; the paper recommends 0.3-0.4 in the text but does not explain why 0.6 was retained on hardware, which is directly relevant to the safety discussion and should be addressed.
  5. [VI.B, Algorithm 1] The whittling algorithm's MakeCut subroutine is solved by a custom gradient-based solver described in Appendix C, but no comparison against a standard convex QP or nonlinear solver is given; a brief validation that the custom solver reliably reaches the same cuts as a reference solver would increase confidence in the polygon decomposition.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the ALIP-derived MPFC cost and dynamics, the big-M foothold constraints, and the hardware timing/performance claims are self-contained; self-citations are historical rather than load-bearing.

full rationale

Walking the paper's derivation chain: the ALIP model is an external modeling assumption (cited [12], [28]) and no parameter of it is fitted to the experiments in this paper; the reset map, step-to-step dynamics, and timing linearization in Sec. III.B-D are derived algebraically from that model. The MPFC cost in Sec. IV.A uses a desired-velocity subspace (period-2 ALIP orbits) constructed explicitly in Appendix B from the ALIP matrices (Eqs. 29-31), not from walking data, so the cost is not a fitted input disguised as a prediction. The foothold constraints (20) are exact big-M encodings of the S3 polygons, and the S3 mask is generated by the explicitly recursive hysteresis rule in Sec. VI.A.3; the high frame-to-frame IoU reported in Fig. 16 is a measurement of that deliberately built-in hysteresis mechanism, which the paper transparently labels as such rather than presenting as an independent derivation. The hand-set thresholds (k_safe=0.7, k_hyst=0.6) and the documented leaf false-positive (Sec. IX.B.2) are correctness/soundness limitations of the safety mask, not circular reasoning: MPFC does not assume the safety conclusion it reports. Self-citations to the authors' precursor [9] support background, motivation, and novelty ('first deployment'), but the central claims of sub-10 ms solve times and successful hardware walking are evidenced by the measurements in this paper (Table II, Fig. 13), so no load-bearing argument reduces to a self-citation. No equation in the paper reuses its own output as an input in a way that would make a claimed prediction equivalent to its assumptions.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claims rest on the ALIP template model from prior work, the linearizations used to build the MIQP, and the hand-tuned S3 safety criteria. The ALIP model is a domain assumption, not derived here; the double-stance approximation is acknowledged and patched; the safety thresholds are ad hoc to the tested terrains. No new physical entities are introduced.

free parameters (8)
  • MPFC cost weights Q, QN, R = Table V: QN = diag[100,100,1,1] (hardware), Q = diag[0.001,0.1,0.01,0.001], R = diag[25,25,0]
    Hand-tuned weights in the MIQP cost; they shape footstep aggressiveness and balance margin. The paper notes QN was set at least 100x Q for the position coordinates.
  • S3 safety threshold k_safe = 0.7
    Pixel safety threshold; chosen to correspond to a roughly 33 degree slope limit. Not derived from the robot's kinematic or dynamic error bounds.
  • S3 hysteresis k_hyst = 0.6 (hardware), 0.4 (simulation)
    Chosen to maximize temporal consistency of the segmentation; sensitivity analysis shows 0.3-0.4 behaves similarly, but no principled selection method is given.
  • Curvature criterion scaling alpha_c and LoG sigma = alpha_c = 5, sigma = 2 px
    Tuned ad hoc to penalize terrain below ledges while not over-penalizing flat ground.
  • Nominal stance durations Tss, Tds and timing bounds = Tss = 0.3 s, Tds = 0.1 s, T in [0.27, 0.33] s
    Nominal gait timing from Cassie's design; the T bounds are hard constraints in the MIQP and limit the linearization error in Eq. (12).
  • Swing foot clearance c = 0.15 m
    Tunable clearance for swing foot trajectories on flat ground; adjusted adaptively in Section V.B.
  • Big-M constant M = 10
    Chosen for numerical stability; must be large enough to relax foothold constraints when a foothold is not selected. Stated in Section IV.C.
  • Erosion safety margin kernel size = 4 px
    Adds a safety margin in the steppability mask to account for foot tracking error and foot length; chosen without a formal error analysis.
assumptions (6)
  • domain assumption The ALIP model with angular momentum about the contact point accurately approximates Cassie's horizontal CoM dynamics.
    Invoked in Section III.A; validated in prior work [12, 28] for robots with heavy legs, but not re-derived or re-validated here.
  • domain assumption The double-stance cross-product term ((pCoP - p-) x m vCoM)_{x,y} is negligible.
    Stated in Section III.B with a parallelism justification; the paper patches lateral error via an instantaneous weight transfer adjustment in Appendix A.
  • standard math First-order linearization of the ALIP flow with respect to remaining stance time T is accurate over the allowed range [0.27, 0.33] s.
    Eq. (12) in Section III.D; the bounds in Table V limit the linearization error, but no error bound is provided.
  • ad hoc to paper Curvature and inclination safety criteria, combined with hysteresis, produce a correct safe/unsafe terrain partition.
    Section VI.A; the thresholds (k_safe = 0.7, k_hyst = 0.6) and scales (alpha_c = 5) are chosen for the tested terrains. The dried-leaves failure in Section IX.B.2 shows the partition can be wrong.
  • domain assumption Elevation map drift correction via median height difference from the stance foot yields an accurate ground height map.
    Section VII.B; relies on the stance foot being in contact with known ground, which can fail during slips or when crossing gaps.
  • standard math The MPC state cost can be expressed as a projection onto a desired-velocity subspace without encoding a footstep pattern.
    Appendix B derives the projection from the s2s dynamics; it assumes the linear model and the period-2 orbit constraint, not fitted to data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Perceptive Mixed-Integer Footstep Control for Underactuated Bipedal Walking on Rough Terrain." pith.science (2026). https://pith.science/paper/Q2YKZ2HT

@misc{pith2026250119391,
  author       = {Pith},
  title        = {Pith review of: Perceptive Mixed-Integer Footstep Control for Underactuated Bipedal Walking on Rough Terrain},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q2YKZ2HT}},
  note         = {Machine review of arXiv:2501.19391}
}
read the original abstract

Traversing rough terrain requires dynamic bipeds to stabilize themselves through foot placement without stepping in unsafe areas. Planning these footsteps online is challenging given non-convexity of the safe terrain, and imperfect perception and state estimation. This paper addresses these challenges with a full-stack perception and control system for achieving underactuated walking on discontinuous terrain. First, we develop model-predictive footstep control (MPFC), a single mixed-integer quadratic program which assumes a convex polygon terrain decomposition to optimize over discrete foothold choice, footstep position, ankle torque, template dynamics, and footstep timing at over 100 Hz. We then propose a novel approach for generating convex polygon terrain decompositions online. Our perception stack decouples safe-terrain classification from fitting planar polygons, generating a temporally consistent terrain segmentation in real time using a single CPU thread. We demonstrate the performance of our perception and control stack through outdoor experiments with the underactuated biped Cassie, achieving state of the art perceptive bipedal walking on discontinuous terrain. Supplemental Video: https://youtu.be/JK16KJXJxi4

Figures

Figures reproduced from arXiv: 2501.19391 by the authors.

Figure 1
Figure 1. The bipedal robot Cassie walks up and down brick [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The perception and control stack proposed in this paper to achieve underactuated walking over discontinuous terrain. Our [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The ALIP model assumes that the robot’s CoM is [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Top: Key MPFC decision variables and constraints for a horizon of 2 stance phases. xc is the current ALIP state, u is ankle torque applied during the current stance phase, x0 is the ALIP state at the end of the current stance phase, and x1 is the ALIP state at the end …
Figure 5
Figure 5. Figure 5: To enforce the planarity assumption of the ALIP, we [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Trajectory from the swing foot position at the beginning [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Pipeline for converting an elevation map of the terrain into a set of convex polygons for planning safe footsteps. Stable [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 9
Figure 9. Figure 9: Walking over ledges with Cassie requires asymmetric [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Example of successfully traversing a random stepping [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: MPFC simulation experiments. Top: we use ground truth terrain information to walk over a 23 cm wide beam, and stairs with a rise of 15 cm and a depth of 27 cm. Bottom: Displaying the elevation map for walking over the same terrain types using S3. Safety margin in S3 r…
Figure 12
Figure 12. Figure 12: Success rates for walking across randomly-generated [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 13
Figure 13. Figure 13: Cassie Walks on unstructured terrain using our proposed perception and control stack, climbing and descending a set [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 15
Figure 15. Figure 15: Detailed profiling of our perception stack, showing the [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]
Figure 14
Figure 14. Figure 14: The environments used to collect data for benchmark [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 16
Figure 16. Figure 16: Offline benchmark of S3 compared to plane segmentation baselines. [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]
Figure 17
Figure 17. Figure 17: Tiles showing the output of each segmentation method for each evaluation environment at 1 second intervals. In the [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]
Figure 18
Figure 18. Figure 18: The final convex decomposition has a similarly shaped [PITH_FULL_IMAGE:figures/full_fig_p016_18.png]
Figure 19
Figure 19. Figure 19: Frame-to-Frame IoU of the S3 terrain segmentation [PITH_FULL_IMAGE:figures/full_fig_p017_19.png]
Figure 20
Figure 20. Figure 20: Our proposed system naturally handles terrain with [PITH_FULL_IMAGE:figures/full_fig_p017_20.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 53 canonical work pages

  1. [1]

    Supplemental Video (short)

    “Supplemental Video (short).” [Online]. Available: https: //youtu.be/qk05xAqjyKQ

  2. [2]

    Supplemental Video (full)

    “Supplemental Video (full).” [Online]. Available: https://youtu. be/JK16KJXJxi4

  3. [3]

    3D dynamic walking on stepping stones with control barrier functions,

    Q. Nguyen, A. Hereid, J. W. Grizzle, A. D. Ames, and K. Sreenath, “3D dynamic walking on stepping stones with control barrier functions,” in 2016 IEEE 55th Conference on Decision and Control (CDC) . Las Vegas, NV , USA: IEEE, Dec. 2016, pp. 827–834

  4. [4]

    Adaptive Step Duration for Precise Foot Placement: Achieving Robust Bipedal Loco- motion on Terrains with Restricted Footholds,

    Z. Xiang, V . Paredes, and A. Hereid, “Adaptive Step Duration for Precise Foot Placement: Achieving Robust Bipedal Loco- motion on Terrains with Restricted Footholds,” Mar. 2024

  5. [5]

    Bipedal Walking on Con- strained Footholds: Momentum Regulation via Vertical COM Control,

    M. Dai, X. Xiong, and A. Ames, “Bipedal Walking on Con- strained Footholds: Momentum Regulation via Vertical COM Control,” in 2022 International Conference on Robotics and Automation (ICRA), May 2022, pp. 10 435–10 441

  6. [6]

    Learning Linear Policies for Robust Bipedal Lo- comotion on Terrains with Varying Slopes,

    L. Krishna, U. A. Mishra, G. A. Castillo, A. Hereid, and S. Kolathaya, “Learning Linear Policies for Robust Bipedal Lo- comotion on Terrains with Varying Slopes,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Sep. 2021, pp. 5159–5164

  7. [7]

    SLIP Walking Over Rough Terrain via H-LIP Stepping and Backstepping-Barrier Function Inspired Quadratic Program,

    X. Xiong and A. Ames, “SLIP Walking Over Rough Terrain via H-LIP Stepping and Backstepping-Barrier Function Inspired Quadratic Program,” IEEE Robotics and Automation Letters , vol. 6, no. 2, pp. 2122–2129, Apr. 2021

  8. [8]

    Topology-Based MPC for Automatic Foot- step Placement and Contact Surface Selection,

    J. Shim, C. Mastalli, T. Corbères, S. Tonneau, V . Ivan, and S. Vijayakumar, “Topology-Based MPC for Automatic Foot- step Placement and Contact Surface Selection,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) . London, United Kingdom: IEEE, May 2023, pp. 12 226–12 232

Show all 55 references
  1. [9]

    Bipedal Walking on Constrained Footholds with MPC Footstep Control,

    B. Acosta and M. Posa, “Bipedal Walking on Constrained Footholds with MPC Footstep Control,” in 2023 IEEE-RAS 22nd International Conference on Humanoid Robots (Hu- manoids), Dec. 2023, pp. 1–8

  2. [10]

    Robust-Locomotion-by-Logic: Perturbation- Resilient Bipedal Locomotion via Signal Temporal Logic Guided Model Predictive Control,

    Z. Gu, Y . Zhao, Y . Chen, R. Guo, J. K. Leestma, G. S. Saw- icki, and Y . Zhao, “Robust-Locomotion-by-Logic: Perturbation- Resilient Bipedal Locomotion via Signal Temporal Logic Guided Model Predictive Control,” Mar. 2024

  3. [11]

    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,” in 2014 IEEE-RAS International Conference on Humanoid Robots , Nov. 2014, pp. 279–286

  4. [12]

    Terrain-Adaptive, ALIP-Based Bipedal Locomotion Controller via Model Predictive Control and Virtual Constraints,

    G. Gibson, O. Dosunmu-Ogunbi, Y . Gong, and J. Grizzle, “Terrain-Adaptive, ALIP-Based Bipedal Locomotion Controller via Model Predictive Control and Virtual Constraints,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Oct. 2022, pp. 6724–6731

  5. [13]

    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. Fernández-López, and C. Semini, “Simultaneous Contact, Gait, and Motion Planning for Robust Multilegged Locomotion via Mixed-Integer Convex Optimization,” IEEE R...

  6. [14]

    3D Hopping in Discontinuous Terrain Using Impulse Planning with Mixed- Integer Strategies,

    N. Fey, R. J. Frei, and P. M. Wensing, “3D Hopping in Discontinuous Terrain Using Impulse Planning with Mixed- Integer Strategies,” IEEE Robotics and Automation Letters , pp. 1–8, 2024

  7. [15]

    Kinodynamic Motion Planning for Multi-Legged Robot Jumping via Mixed-Integer Convex Program,

    Y . Ding, C. Li, and H.-W. Park, “Kinodynamic Motion Planning for Multi-Legged Robot Jumping via Mixed-Integer Convex Program,” in 2020 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS) , Oct. 2020, pp. 3998–4005

  8. [16]

    Per- ceptive Locomotion through Whole-Body MPC and Optimal Region Selection,

    T. Corbères, C. Mastalli, W. Merkt, I. Havoutis, M. Fallon, N. Mansard, T. Flayols, S. Vijayakumar, and S. Tonneau, “Per- ceptive Locomotion through Whole-Body MPC and Optimal Region Selection,” May 2023

  9. [17]

    Perceptive Locomotion through Nonlinear Model Predictive Control,

    R. Grandia, F. Jenelten, S. Yang, F. Farshidian, and M. Hutter, “Perceptive Locomotion through Nonlinear Model Predictive Control,” Aug. 2022

  10. [18]

    Approximate convex decompo- sition of polygons,

    J.-M. Lien and N. M. Amato, “Approximate convex decompo- sition of polygons,” Computational Geometry , vol. 35, no. 1, pp. 100–123, Aug. 2006

  11. [19]

    A Fast, Autonomous, Bipedal Walking Behavior over Rapid Regions,

    D. Calvert, B. Mishra, S. McCrory, S. Bertrand, R. Griffin, and J. Pratt, “A Fast, Autonomous, Bipedal Walking Behavior over Rapid Regions,” in 2022 IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids) , Nov. 2022, pp. 24–31

  12. [20]

    Griffin, G

    R. Griffin, G. Wiedebach, S. McCrory, S. Bertrand, I. Lee, and J. Pratt, Footstep Planning for Autonomous Walking Over Rough Terrain, Jul. 2019

  13. [21]

    Biped walking pattern generation by using preview control of zero-moment point,

    S. Kajita, F. Kanehiro, K. Kaneko, K. Fujiwara, K. Harada, K. Yokoi, and H. Hirukawa, “Biped walking pattern generation by using preview control of zero-moment point,” in 2003 IEEE International Conference on Robotics and Automation (Cat. No.03CH37422), vol. 2, 2003, pp. 1620–...

  14. [22]

    SL1M: Sparse L1-norm Minimization for contact planning on uneven terrain,

    S. Tonneau, D. Song, P. Fernbach, N. Mansard, M. Taix, and A. Del Prete, “SL1M: Sparse L1-norm Minimization for contact planning on uneven terrain,” in 2020 IEEE International Conference on Robotics and Automation (ICRA). Paris, France: IEEE, May 2020, pp. 6604–6610

  15. [23]

    Solving Footstep Planning as a Feasibility Problem Using L1-Norm Minimization,

    D. Song, P. Fernbach, T. Flayols, A. D. Prete, N. Mansard, S. Tonneau, and Y . J. Kim, “Solving Footstep Planning as a Feasibility Problem Using L1-Norm Minimization,” IEEE Robotics and Automation Letters , vol. 6, no. 3, pp. 5961–5968, Jul. 2021

  16. [24]

    Real-time Footstep Planning and Control of the Solo Quadruped Robot in 3D Environments,

    F. Risbourg, T. Corbères, P.-A. Léziart, T. Flayols, N. Mansard, and S. Tonneau, “Real-time Footstep Planning and Control of the Solo Quadruped Robot in 3D Environments,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Oct. 2022, pp. 12 950–12 956

  17. [25]

    The 3D linear inverted pendulum mode: A simple modeling for a biped walking pattern generation,

    S. Kajita, F. Kanehiro, K. Kaneko, K. Yokoi, and H. Hirukawa, “The 3D linear inverted pendulum mode: A simple modeling for a biped walking pattern generation,” in Proceedings 2001 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), vol. 1, Oct. 2001, pp...

  18. [26]

    3-D Underactuated Bipedal Walking via H-LIP Based Gait Synthesis and Stepping Stabilization,

    X. Xiong and A. Ames, “3-D Underactuated Bipedal Walking via H-LIP Based Gait Synthesis and Stepping Stabilization,” IEEE Transactions on Robotics , vol. 38, no. 4, pp. 2405–2425, 19 Aug. 2022

  19. [27]

    Reachability Aware Capture Regions with Time Adjustment and Cross-Over for Step Recovery,

    R. Griffin, J. Foster, S. Pasano, B. Shrewsbury, and S. Bertrand, “Reachability Aware Capture Regions with Time Adjustment and Cross-Over for Step Recovery,” in 2023 IEEE-RAS 22nd International Conference on Humanoid Robots (Humanoids) , Dec. 2023, pp. 1–8

  20. [28]

    One-Step Ahead Prediction of Angular Momentum about the Contact Point for Control of Bipedal Locomotion: Validation in a LIP-inspired Controller,

    Y . Gong and J. Grizzle, “One-Step Ahead Prediction of Angular Momentum about the Contact Point for Control of Bipedal Locomotion: Validation in a LIP-inspired Controller,” in 2021 IEEE International Conference on Robotics and Automation (ICRA), May 2021, pp. 2832–2838

  21. [29]

    Stair Climbing using the Angular Momentum Linear Inverted Pendulum Model and Model Predictive Control,

    O. Dosunmu-Ogunbi, A. Shrivastava, G. Gibson, and J. W. Grizzle, “Stair Climbing using the Angular Momentum Linear Inverted Pendulum Model and Model Predictive Control,” Jul. 2023

  22. [30]

    Continuous humanoid locomotion over uneven terrain using stereo fusion,

    M. F. Fallon, P. Marion, R. Deits, T. Whelan, M. Antone, J. Mc- Donald, and R. Tedrake, “Continuous humanoid locomotion over uneven terrain using stereo fusion,” in2015 IEEE-RAS 15th International Conference on Humanoid Robots (Humanoids) , Nov. 2015, pp. 881–888

  23. [31]

    Probabilistic Terrain Mapping for Mobile Robots With Uncertain Localization,

    P. Fankhauser, M. Bloesch, and M. Hutter, “Probabilistic Terrain Mapping for Mobile Robots With Uncertain Localization,” IEEE Robotics and Automation Letters , vol. 3, no. 4, pp. 3019– 3026, Oct. 2018

  24. [32]

    Elevation Mapping for Locomotion and Navi- gation using GPU,

    T. Miki, L. Wellhausen, R. Grandia, F. Jenelten, T. Homberger, and M. Hutter, “Elevation Mapping for Locomotion and Navi- gation using GPU,” Apr. 2022

  25. [33]

    Plane Seg – Robustly and Efficiently Extracting Contact Regions from Depth Data,

    M. Fallon and M. Antone, “Plane Seg – Robustly and Efficiently Extracting Contact Regions from Depth Data,”

  26. [34]

    Vilens: Visual, inertial, lidar, and leg odometry for all-terrain legged robots,

    D. Wisth, M. Camurri, and M. Fallon, “Vilens: Visual, inertial, lidar, and leg odometry for all-terrain legged robots,” IEEE Transactions on Robotics , vol. 39, no. 1, pp. 309–326, 2023

  27. [35]

    Real-Time Polygonal Semantic Mapping for Humanoid Robot Stair Climbing,

    T. Bin, J. Yao, T. Lun Lam, and T. Zhang, “Real-Time Polygonal Semantic Mapping for Humanoid Robot Stair Climbing,” in 2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids), Nov. 2024

  28. [36]

    Bipedal navigation planning over rough terrain using traversability models,

    S. McCrory, B. Mishra, R. Griffin, J. Pratt, and H. E. Sevil, “Bipedal navigation planning over rough terrain using traversability models,” in SoutheastCon 2023, 2023, pp. 89–95

  29. [37]

    TA- MOLS: Terrain-Aware Motion Optimization for Legged Sys- tems,

    F. Jenelten, R. Grandia, F. Farshidian, and M. Hutter, “TA- MOLS: Terrain-Aware Motion Optimization for Legged Sys- tems,” IEEE Transactions on Robotics, vol. 38, no. 6, pp. 3395– 3413, Dec. 2022

  30. [38]

    Blind Bipedal Stair Traversal via Sim-to-Real Reinforcement Learning,

    J. Siekmann, K. Green, J. Warila, A. Fern, and J. Hurst, “Blind Bipedal Stair Traversal via Sim-to-Real Reinforcement Learning,” in Robotics: Science and Systems XVII . Robotics: Science and Systems Foundation, Jul. 2021

  31. [39]

    Sim-to-Real Learning of Footstep- Constrained Bipedal Dynamic Walking,

    H. Duan, A. Malik, J. Dao, A. Saxena, K. Green, J. Siekmann, A. Fern, and J. Hurst, “Sim-to-Real Learning of Footstep- Constrained Bipedal Dynamic Walking,” in 2022 International Conference on Robotics and Automation (ICRA), May 2022, pp. 10 428–10 434

  32. [40]

    Learning vision-based bipedal locomo- tion for challenging terrain,

    H. Duan, B. Pandit, M. S. Gadde, B. J. van Marum, J. Dao, C. Kim, and A. Fern, “Learning vision-based bipedal locomo- tion for challenging terrain,” arXiv preprint arXiv:2309.14594 , 2023

  33. [41]

    Dtc: Deep tracking control,

    F. Jenelten, J. He, F. Farshidian, and M. Hutter, “Dtc: Deep tracking control,” Science Robotics, vol. 9, no. 86, p. eadh5401, 2024

  34. [42]

    Learning generic and dynamic locomotion of humanoids across discrete terrains,

    S. Yu, N. Perera, D. Marew, and D. Kim, “Learning generic and dynamic locomotion of humanoids across discrete terrains,” arXiv preprint arXiv:2405.17227 , 2024

  35. [43]

    Starting on the Right Foot with Reinforcement Learning

    Boston Dynamics, “Starting on the Right Foot with Reinforcement Learning.” [On- line]. Available: https://bostondynamics.com/blog/ starting-on-the-right-foot-with-reinforcement-learning/

  36. [44]

    Ogata, Modern Control Engineering, 4th ed

    K. Ogata, Modern Control Engineering, 4th ed. USA: Prentice Hall PTR, 2001

  37. [45]

    Three- dimensional bipedal walking control using divergent component of motion,

    J. Englsberger, C. Ott, and A. Albu-Schäffer, “Three- dimensional bipedal walking control using divergent component of motion,” in 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems , 2013, pp. 2600–2607

  38. [46]

    Walking Control Based on Step Timing Adaptation,

    M. Khadiv, A. Herzog, S. A. A. Moosavian, and L. Righetti, “Walking Control Based on Step Timing Adaptation,” Mar. 2020

  39. [47]

    Generation of dynamic hu- manoid behaviors through task-space control with conic opti- mization,

    P. M. Wensing and D. E. Orin, “Generation of dynamic hu- manoid behaviors through task-space control with conic opti- mization,” in 2013 IEEE International Conference on Robotics and Automation, May 2013, pp. 3103–3109

  40. [48]

    A training algorithm for optimal margin classifiers,

    B. E. Boser, I. M. Guyon, and V . N. Vapnik, “A training algorithm for optimal margin classifiers,” in Proceedings of the fifth annual workshop on Computational learning theory , 1992, pp. 144–152

  41. [49]

    LCM: Lightweight Communications and Marshalling,

    A. S. Huang, E. Olson, and D. C. Moore, “LCM: Lightweight Communications and Marshalling,” in 2010 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems, Oct. 2010, pp. 4057–4062

  42. [50]

    Drake: Model-Based Design and Verification for Robotics,

    Russ Tedrake and the Drake Development Team, “Drake: Model-Based Design and Verification for Robotics,” 2019

  43. [51]

    FCCQP: A Whole Body Control QP Solver with Full Friction Cones

    B. Acosta, “FCCQP: A Whole Body Control QP Solver with Full Friction Cones.” [Online]. Available: https://github.com/ Brian-Acosta/fcc_qp/

  44. [52]

    Contact-aided invariant extended Kalman filtering for robot state estimation,

    R. Hartley, M. Ghaffari, R. M. Eustice, and J. W. Grizzle, “Contact-aided invariant extended Kalman filtering for robot state estimation,” The International Journal of Robotics Re- search, vol. 39, no. 4, pp. 402–430, Mar. 2020

  45. [53]

    Efficient ransac for point- cloud shape detection,

    R. Schnabel, R. Wahl, and R. Klein, “Efficient ransac for point- cloud shape detection,” Computer Graphics Forum , vol. 26, 2007

  46. [54]

    Navier-stokes, fluid dynamics, and image and video inpainting,

    M. Bertalmio, A. Bertozzi, and G. Sapiro, “Navier-stokes, fluid dynamics, and image and video inpainting,” in Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001 , vol. 1, 2001, pp. I–I. APPENDIX A. Lateral Reset Map A...

  47. [2019]

    Available: https://github.com/ori-drs/plane_seg

    [Online]. Available: https://github.com/ori-drs/plane_seg

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.