Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

TRG-planner: Traversal Risk Graph-Based Path Planning in Unstructured Environments for Safe and Efficient Navigation

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read TRG-planner encodes rough terrain as a risk-weighted graph and finds safer, faster paths than A*, PRM*, or T-Hybrid.

desk verdict Solid field robotics with a genuinely useful graph representation, but Eq. (4) leaves the PCA eigenvector orientation unspecified, which is a real reproducibility gap in the central risk definition. read the letter →

arxiv 2501.01806 v1 pith:6PDGI5CW submitted 2025-01-03 cs.RO cs.SYeess.SY

Dongkyu Lee , I Made Aswin Nahrendra , Minho Oh , Byeongho Yu , Hyun Myung This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords PathplanningTraversabilityUnstructuredenvironmentsLeggedrobotsFieldroboticsTraversalRiskGraphGraph-based
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

The paper proposes TRG-planner, a path-planning method for legged robots in unstructured terrain. Its central claim is that representing the environment as a graph whose edges carry a direction-dependent traversal risk—computed from local slope steepness relative to travel direction—lets a planner find routes that are both safer and shorter than those from A*, PRM*, and T-Hybrid. In 50 m × 50 m simulations with random start-goal pairs, the balanced variant achieves higher travel success and lower path risk than all baselines, with planning times around 3 ms in the longest scenario, roughly an order of magnitude faster than the fastest baseline. The same planner successfully guided a quadruped over a mountain, a one-sided mound, and the ICRA 2023 Quadruped Robot Challenge arena. The method is intended as a global planner that can combine with downstream local optimizers or semantic traversability estimates.

What carries the argument

The central object is the Traversal Risk Graph (TRG), a graph G = (V, E) whose nodes are stance patches and whose edges are risk-weighted path candidates. Nodes are sampled outward from the current robot position in a wavefront manner, so only reachable terrain enters the graph; each node stores its 3D position, a validity flag combining geometric stability (height variation below a threshold within a robot-sized inscribed circle) and reachability (at least one incident edge), and its incident edges. Edges are wired between nearby nodes only if the local terrain between them is not too rough or steep; each edge weight is built from a PCA fit of the terrain ellipse around the path, with longitudinal and lateral risk components Rξ_lon and Rξ_lat computed as negative inner products of the principal eigenvectors with gravity. This makes the risk direction-aware: the same slope costs differently when approached from different directions. The graph is extracted locally for updates, expanded from frontier nodes, and integrated back into a global graph; the planning cost in Eq. (7) combines edge distance with the risk weight, enabling A* to trade off distance and safety through the safety factor Γ.

What would settle it

On the elevation map used in the Long scenario, pick an edge that crosses a slope, evaluate the cost in Eq. (7) twice: once with the principal eigenvector from a standard PCA routine and once with that eigenvector negated. If the resulting route or total cost differs, the risk weight depends on an arbitrary sign convention rather than the terrain geometry.

Watch

Extended reading notes

Core claim

The central discovery is that relative traversal risk, not just terrain stability, should be the unit of path cost in unstructured environments. TRG-planner builds a Traversal Risk Graph (TRG): nodes are circular terrain patches that the robot can stand on and reach from its current pose, and edges connect nodes whose intervening region is not too steep and whose local height variation is below a threshold. Each edge carries a weight equal to a convex combination of the negative inner products of the two principal PCA direction vectors of the terrain ellipse with gravity, so a path climbing or traversing a slope accumulates risk proportional to its alignment with the fall line. The planner then runs A* on this graph with a cost that sums Euclidean distance plus the risk weight scaled by a safety factor Γ. The paper reports that this representation allows the robot to consistently find paths with lower risk weight, higher travel success, and shorter planning time than conventional methods, and that it worked on a real quadruped in three unstructured field environments.

Load-bearing premise

The planner's risk value on every edge is built from PCA eigenvectors that are defined only up to a sign, and the paper never fixes their orientation relative to the travel direction, so flipping a sign flips the edge's risk weight and changes the route the cost function produces.

Editorial extensions

If this is right

  • In the 50 m × 50 m simulation with 100 random start-goal pairs per scenario, the balanced Γ = 3.0 variant beats A*, PRM*, and T-Hybrid in travel success rate for short, medium, and long scenarios, and matches or beats them in path risk weight W.
  • Planning time is roughly an order of magnitude lower: about 3.00 ms for the long scenario versus 16.89 ms for PRM* and 482.28 ms for T-Hybrid, with no map-preprocessing phase beyond a 4.01 s graph initialization.
  • The direction-aware risk lets the planner select safe entry directions onto slopes, so the robot climbs a mound only on the side it can actually ascend, and completes the QRC arena course without falling.
  • The method's behavior can be tuned continuously between distance-minimizing (Γ = 1.0) and risk-minimizing (Γ = 10.0), with the balanced choice producing the best combined travel deviation and success rate.

Reading between the lines

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

  • Because edge weights are geometric, they can be combined additively with semantic or learned traversability costs, so the same graph could serve as a global cost skeleton for planners that use image-based terrain classification or learned locomotion costs.
  • The wavefront construction with frontier-based expansion is naturally suited to streaming elevation maps; a formal incremental variant could update the global graph without rebuilding it, which would reduce replanning latency further as map horizons grow.
  • The same cost structure could generalize from legged robots to wheeled or tracked platforms by replacing the gravity inner product with a platform-specific rollover or slip risk model, without changing the graph machinery.
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 / 5 minor

Summary. The paper presents TRG-planner, a global path-planning method for unstructured terrain. The environment is represented as a Traversal Risk Graph whose nodes encode local standability/reachability and whose edges carry a traversal-risk weight derived from PCA of the local height map. The graph is built by wavefront sampling and updated hierarchically, and planning is performed by an A*-style search over a cost that combines Euclidean distance with the risk weight. The authors claim, based on simulation against A*, PRM*, and T-Hybrid and on real-world quadruped experiments (including the ICRA 2023 Quadruped Robot Challenge), that TRG-planner gives higher travel success, lower path risk, shorter paths, and much faster planning.

Significance. If the definitions were fully specified, the contribution would be of practical interest to field robotics: a real-time risk-aware global planner validated on a legged platform and in a competitive setting. The hierarchical graph construction and the wavefront propagation idea are reasonable engineering contributions, and the reported planning times are notably low. However, the central risk definition in Eq. (4) has a sign ambiguity that makes the planner's objective not fully defined as written, and the main safety metric in Eq. (8) is computed from the very edge weights that the planner minimizes. These issues must be resolved before the safety and efficiency claims can be assessed. The paper is currently not reproducible from the text alone.

major comments (4)
  1. [Section III.A.2, Eq. (4)] The sign of the risk weight is not defined. The quantities Rξ_lon and Rξ_lat are defined as negative inner products of PCA eigenvectors with g = [0,0,-1]. PCA eigenvectors are determined only up to a sign, and the text does not specify any orientation rule relating êξ_lon to the travel direction from vi to vj or to the uphill direction. Under a literal reading, a standard PCA routine may return êξ_lon pointing in either direction, so w_ij can flip sign, and the graph can have negative or inconsistent edge weights. This is not a cosmetic issue: the cost update in Eq. (7) uses (Γw_{i+1,i} + 1), which becomes negative for w < -1/Γ, breaking the admissibility of the heuristic J and potentially causing the planner to prefer unrealistic paths. The description of TRG as an undirected graph also conflicts with the claimed direction-aware risk, since a single scalar w_ij cannot represent different risk for traversal from vi to vj versus from vj to vi. The authors must specify an orientation rule for the eigenvectors, clarify whether risk is symmetric or directed, and show that the resulting edge weights are nonnegative.
  2. [Section IV.B, Eq. (8)] The normalized path risk W is not an independent safety measure for the proposed planner. W is computed as the average of the TRG edge weights w_{i+1,i} along the planned path, and those same weights are exactly the quantities minimized by the TRG-planner cost function in Eq. (7). Consequently, a comparison of W between TRG-planner and the baselines is biased in favor of TRG-planner by construction, especially because the baselines never consider w in their objectives. The travel success rate Strav and the deviation metric T are more independent, but W is presented as a main safety metric in Tables II and IV and in Section V. The authors should replace W with an externally defined terrain-risk measure (e.g., slope/roughness statistics along the traversed trajectory, or measured body orientation/contact events) or at least report the raw terrain properties of the actually traveled paths.
  3. [Section V.A, Table II] The two risk-related parameters γ and Γ are tuned on the same test scenarios used for the final comparison, and the balanced strategy Γ=3.0 is selected post hoc from three tested values (1.0, 3.0, 10.0). This is a form of fitting to the test set, and it weakens the generality of the claim that the balanced strategy is best. In addition, Tables II-IV report only point estimates for metrics such as Spath, Strav, Lpath, W, and T, even though each scenario uses 100 randomly generated start-goal pairs; no error bars, confidence intervals, or statistical significance tests are provided. The authors should report variances or confidence intervals and should validate the parameter choice on a held-out set or via sensitivity analysis.
  4. [Section III.B and III.C] The graph construction and management rely on several informal terms that should be made precise for reproducibility: the 'vicinity' of the reference node and the radius rexp are clear enough, but the 'area covered by G' (QG in Eq. (5)) is not formally defined, and the frontier condition ui ∉ QG depends on this undefined set. Also, the statement that nodes are sampled 'following a uniform distribution on a circle' with radius rexp means the nodes lie on a ring rather than in a disk; this should be stated explicitly, since it affects graph connectivity and coverage.
minor comments (5)
  1. [Section IV.B] The definition T = 0 when negative is ad hoc; the authors should explain why a negative T arises and why it is truncated rather than treated as a meaningful value.
  2. [Section V.B, Table III] The comparison with PRM* uses 'the same number of samples as the TRG nodes,' but the initialization time and roadmap quality of PRM* depend strongly on the sampling strategy and connection radius; the comparison would be clearer if the PRM* parameters were reported and justified.
  3. [Section III.A.2] In Eq. (4), the notation Rξ_dir is defined but not used afterward; the definitions of Rξ_lon and Rξ_lat would be easier to follow if the authors explicitly wrote out the components of êξ_lon and êξ_lat and the sign convention for uphill versus downhill.
  4. [Section IV.A.1] The simulation environment is described as '50 m × 50 m × 6.9 m,' but Fig. 6(b) labels show '6.9 0.0' and the environment appears to be a height map; please clarify whether 6.9 m is the elevation range or the vertical extent of the map.
  5. [Section II.B] The related work on traversability-aware planning is discussed adequately, but the comparison with the STEP planner [22] and SMUG planner [23] is only brief; since those methods also address risk-aware global planning, a paragraph contrasting their risk formulations with the TRG formulation would help.

Circularity Check

1 steps flagged · score 6.0 of 10

W safety metric reduces to the planner's own objective: Eq. (8) averages the TRG edge weights that Eq. (7) minimizes, making W-based safety comparisons for TRG-planner circular, while travel-success and real-world evidence remain independent.

  1. self definitional [Section IV.B, Eq. (8); Section III.D, Eq. (7)]
    "W is derived from the weights of TRG edges along the planned path, to reflect the safety level of the path. ... W = 1/Lpath Σ_{i=1}^{n-1} w_{i+1,i} ... C(vi+1) = C(vi) + di+1,i(Γwi+1,i + 1)."

    The safety metric W is defined as the average of the same edge weights w_{i+1,i} that TRG-planner explicitly minimizes through the cost function in Eq. (7). Reporting low W for TRG-planner in Tables II and IV is therefore not an independent test of safety; it is the planner's own objective restated as an evaluation metric. For the proposed method the comparison is self-definitional, although W remains an external metric for A*, PRM*, and T-Hybrid. This does not invalidate the independent Strav, T, planning-time, or real-world results, but it does make the W-based safety claim circular.

full rationale

The central derivation of TRG edge weights, graph construction, and A* cost formulation is self-contained: no load-bearing self-citation, imported uniqueness theorem, or ansatz smuggled in by citation was found. The authors' own DreamWaQ controller is used as an external locomotion component and does not justify the planner's novelty. The main circularity is evaluative rather than derivational: Eq. (8) defines the normalized path risk W as the average of the TRG edge weights w_{i+1,i}, and Eq. (7) minimizes exactly those weights (scaled by Γ and distance). Thus Table IV's W comparisons for TRG-planner are the planner's cost function reported as a safety metric, so the W-based superiority claim is forced by construction. Independent support for safety remains through travel success rate Strav, path deviation T, and the real-world demonstrations, so the paper is only partially circular. Separately, Eq. (4) leaves the orientation (sign) of the PCA eigenvectors unspecified, which can flip Rξ_lon and Rξ_lat and hence edge weights; this is a reproducibility/correctness flaw rather than a circularity, since it does not make the derivation equivalent to its inputs.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces the Traversal Risk Graph as a new data structure, but no new physical entities such as particles, forces, or dimensions. The free parameters are all empirically set or robot-specific thresholds. The axioms are domain assumptions about terrain representation and cost design; none are standard mathematical axioms beyond the use of PCA and A*.

free parameters (5)
  • gamma (risk ratio) = 0.2
    Ratio of longitudinal to latitudinal risk in edge weight Eq. (4); empirically set, not derived.
  • Gamma (safety factor) = 3.0 (balanced strategy)
    Safety factor in cost function Eq. (7); tuned by comparing 1.0, 3.0, and 10.0 on the test scenarios, selecting the balanced value post hoc.
  • hmax (height threshold) = 0.16 m
    Maximum height deviation for node stability and edge slope filtering; a hand-set, robot-specific threshold.
  • rexp (node expansion radius) = 0.6 m
    Sampling radius for new nodes; a design choice that controls graph density and edge length.
  • rrobot (robot radius) = 0.3 m
    Inscribed circle radius representing the robot footprint; physical parameter but used in the slope threshold Eq. (3).
assumptions (4)
  • domain assumption PCA eigenvectors of the terrain height map in the edge ellipse provide meaningful longitudinal and latitudinal risk directions, and their sign can be treated consistently.
    Used in Eq. (4) to define risk weights; the paper does not specify how eigenvector signs are oriented relative to the path direction.
  • domain assumption The inscribed circle of radius rrobot and the median-height threshold hmax adequately represent robot stability and obstacle clearance.
    Defines node validity g(ci) in Eq. (1); ignores robot dynamics, foothold placement, and terrain roughness beyond the height threshold.
  • domain assumption The linear weighted sum of distance and risk in Eq. (7) captures the safe-and-short trade-off across different terrains.
    Assumes a scalar cost with fixed Gamma is sufficient; no calibration procedure is provided and the balance factor is empirically chosen.
  • domain assumption The slope bound in Eq. (3), arctan(hmax/rrobot), is a valid edge-steepness filter.
    Heuristic threshold linking step height to robot radius; no kinematic derivation is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TRG-planner: Traversal Risk Graph-Based Path Planning in Unstructured Environments for Safe and Efficient Navigation." pith.science (2026). https://pith.science/paper/6PDGI5CW

@misc{pith2026250101806,
  author       = {Pith},
  title        = {Pith review of: TRG-planner: Traversal Risk Graph-Based Path Planning in Unstructured Environments for Safe and Efficient Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6PDGI5CW}},
  note         = {Machine review of arXiv:2501.01806}
}
read the original abstract

Unstructured environments such as mountains, caves, construction sites, or disaster areas are challenging for autonomous navigation because of terrain irregularities. In particular, it is crucial to plan a path to avoid risky terrain and reach the goal quickly and safely. In this paper, we propose a method for safe and distance-efficient path planning, leveraging Traversal Risk Graph (TRG), a novel graph representation that takes into account geometric traversability of the terrain. TRG nodes represent stability and reachability of the terrain, while edges represent relative traversal risk-weighted path candidates. Additionally, TRG is constructed in a wavefront propagation manner and managed hierarchically, enabling real-time planning even in large-scale environments. Lastly, we formulate a graph optimization problem on TRG that leads the robot to navigate by prioritizing both safe and short paths. Our approach demonstrated superior safety, distance efficiency, and fast processing time compared to the conventional methods. It was also validated in several real-world experiments using a quadrupedal robot. Notably, TRG-planner contributed as the global path planner of an autonomous navigation framework for the DreamSTEP team, which won the Quadruped Robot Challenge at ICRA 2023. The project page is available at https://trg-planner.github.io .

Figures

Figures reproduced from arXiv: 2501.01806 by the authors.

Figure 1
Figure 1. (L-R): Our quadruped robot autonomously navigates through a harsh [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our proposed path planner framework called [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The edge area is approximated to an ellipse plane [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Example of hierarchical graph management. (a) The robot navigates [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: (a) Our quadruped robot, utilized in real-world environments, is [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the success rate of path planning ( [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Qualitative comparison results for five sequences of the path planning simulation. Each planned path is represented by a different color for [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: TRG-planner generated paths from a real mountainous environment. The first column shows the overall map of the environment and the robot’s [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: The mound has different slopes on the left ( [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: TRG-planner generated the global path from the QRC competition arena (left column) and (a-e) the detailed path planning process. The robot [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Context-Aware Risk Estimation in Home Environments: A Probabilistic Framework for Service Robots

    cs.RO 2025-08 reject novelty 5.0 of 10

    A semantic graph framework propagates risk scores derived from a national accident database across spatial object relations, reporting 75% binary risk detection accuracy on 20 human-annotated NYU V2 home images.

Reference graph

Works this paper leans on

39 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    FAR Planner: Fast, attemptable route planner using dynamic visibility update,

    F. Yang, C. Cao, H. Zhu, J. Oh, and J. Zhang, “FAR Planner: Fast, attemptable route planner using dynamic visibility update,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2022, pp. 9–16

    work page 2022

  2. [2]

    Learning forward dynamics model and informed trajectory sampler for safe quadruped navigation,

    Y . Kim, C. Kim, and J. Hwangbo, “Learning forward dynamics model and informed trajectory sampler for safe quadruped navigation,” in Robot. Sci. Syst. , 2022, doi: https://doi.org/10.15607/rss.2022.xviii.069

    work page doi:10.15607/rss.2022.xviii.069 2022

  3. [3]

    A global-local coupling two-stage path planning method for mobile robots,

    Z. Jian, S. Zhang, S. Chen, Z. Nan, and N. Zheng, “A global-local coupling two-stage path planning method for mobile robots,” IEEE Robot. Automat. Lett. , vol. 6, no. 3, pp. 5349–5356, 2021

    work page 2021

  4. [4]

    RT-Mover: a rough terrain mobile robot with a simple leg–wheel hybrid mechanism,

    S. Nakajima, “RT-Mover: a rough terrain mobile robot with a simple leg–wheel hybrid mechanism,” Int. J. Robot. Res. , vol. 30, no. 13, pp. 1609–1626, 2011

    work page 2011

  5. [5]

    High- flexibility locomotion and whole-torso control for a wheel-legged robot on challenging terrain,

    K. Xu, S. Wang, X. Wang, J. Wang, Z. Chen, and D. Liu, “High- flexibility locomotion and whole-torso control for a wheel-legged robot on challenging terrain,” in Proc. IEEE Int. Conf. Robot. Automat. , 2020, pp. 10 372–10 377. (a) (b) (c) (e) (d) (a) (b) (c) (d) (e) Fig. 11. TRG-planner generated the global path from the QRC competition arena (left column...

    work page 2020

  6. [6]

    DreamWaQ: Learning robust quadrupedal locomotion with implicit terrain imagination via deep reinforcement learning,

    I. M. A. Nahrendra, B. Yu, and H. Myung, “DreamWaQ: Learning robust quadrupedal locomotion with implicit terrain imagination via deep reinforcement learning,” in Proc. IEEE Int. Conf. Robot. Au- tomat., 2023, pp. 5078–5084

    2023

  7. [7]

    ViNL: Visual navigation and locomotion over obstacles,

    S. Kareer, N. Yokoyama, D. Batra, S. Ha, and J. Truong, “ViNL: Visual navigation and locomotion over obstacles,” in Proc. IEEE Int. Conf. Robot. Automat. , 2023, pp. 2018–2024

    work page 2023

  8. [8]

    Learning quadrupedal locomotion on deformable terrain,

    S. Choi, G. Ji, J. Park, H. Kim, J. Mun, J. H. Lee, and J. Hwangbo, “Learning quadrupedal locomotion on deformable terrain,” Sci. Robot., vol. 8, no. 74, p. eade2256, 2023

    2023

Show all 39 references
  1. [9]

    Locomotion policy guided traversability learning using volumetric representations of complex environments,

    J. Frey, D. Hoeller, S. Khattak, and M. Hutter, “Locomotion policy guided traversability learning using volumetric representations of complex environments,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst., 2022, pp. 5722–5729

    2022

  2. [10]

    Elevation mapping for locomotion and navigation using GPU,

    T. Miki, L. Wellhausen, R. Grandia, F. Jenelten, T. Homberger, and M. Hutter, “Elevation mapping for locomotion and navigation using GPU,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2022, pp. 2273–2280

    2022

  3. [11]

    Bayesian generalized kernel inference for terrain traversability mapping,

    T. Shan, J. Wang, B. Englot, and K. Doherty, “Bayesian generalized kernel inference for terrain traversability mapping,” in Conf. Robot Learning, 2018, pp. 829–838

    2018

  4. [12]

    Using occupancy grids for mobile robot perception and navigation,

    A. Elfes, “Using occupancy grids for mobile robot perception and navigation,” Computer, vol. 22, no. 6, pp. 46–57, 1989

    1989

  5. [13]

    OctoMap: An efficient probabilistic 3D mapping framework based on Octrees,

    A. Hornung, K. M. Wurm, M. Bennewitz, C. Stachniss, and W. Bur- gard, “OctoMap: An efficient probabilistic 3D mapping framework based on Octrees,” Auton. Robot. , vol. 34, pp. 189–206, 2013

    2013

  6. [14]

    Driving on point clouds: Motion planning, trajectory optimization, and terrain assess- ment in generic nonplanar environments,

    P. Kr ¨usi, P. Furgale, M. Bosse, and R. Siegwart, “Driving on point clouds: Motion planning, trajectory optimization, and terrain assess- ment in generic nonplanar environments,” J. Field Robot. , vol. 34, no. 5, pp. 940–984, 2017

    2017

  7. [15]

    V oxblox: Incremental 3D Euclidean signed distance fields for on- board MA V planning,

    H. Oleynikova, Z. Taylor, M. Fehr, R. Siegwart, and J. Nieto, “V oxblox: Incremental 3D Euclidean signed distance fields for on- board MA V planning,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst., 2017, pp. 1366–1373

    2017

  8. [16]

    Robot-centric elevation mapping with uncertainty estimates,

    P. Fankhauser, M. Bloesch, C. Gehring, M. Hutter, and R. Siegwart, “Robot-centric elevation mapping with uncertainty estimates,” in Mob. Serv. Robot., 2014, pp. 433–440

    2014

  9. [17]

    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 Robot. Automat. Lett. , vol. 3, no. 4, pp. 3019–3026, 2018

    2018

  10. [18]

    TRA VEL: Traversable ground and above- ground object segmentation using graph representation of 3D LiDAR scans,

    M. Oh, E. Jung, H. Lim, W. Song, S. Hu, E. M. Lee, J. Park, J. Kim, J. Lee, and H. Myung, “TRA VEL: Traversable ground and above- ground object segmentation using graph representation of 3D LiDAR scans,” IEEE Robot. Automat. Lett. , pp. 7255–7262, 2022

    2022

  11. [19]

    GaitMesh: controller- aware navigation meshes for long-range legged locomotion planning in multi-layered environments,

    M. Brandao, O. B. Aladag, and I. Havoutis, “GaitMesh: controller- aware navigation meshes for long-range legged locomotion planning in multi-layered environments,” IEEE Robot. Automat. Lett. , vol. 5, no. 2, pp. 3596–3603, 2020

    2020

  12. [20]

    Navigation planning for legged robots in challenging terrain,

    M. Wermelinger, P. Fankhauser, R. Diethelm, P. Kr ¨usi, R. Siegwart, and M. Hutter, “Navigation planning for legged robots in challenging terrain,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2016, pp. 1184–1189

    2016

  13. [21]

    Fast global motion planning for dynamic legged robots,

    J. Norby and A. M. Johnson, “Fast global motion planning for dynamic legged robots,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2020, pp. 3829–3836

    2020

  14. [22]

    STEP: Stochastic traversability evaluation and planning for risk-aware off-road navigation,

    D. D. Fan, K. Otsu, Y . Kubo, A. Dixit, J. Burdick, and A.-A. Agha- Mohammadi, “STEP: Stochastic traversability evaluation and planning for risk-aware off-road navigation,” in Robot. Sci. Syst. , 2022, doi: https://doi.org/10.15607/rss.2021.xvii.021

    2022 doi

  15. [23]

    SMUG Planner: A safe multi-goal planner for mobile robots in challenging environments,

    C. Chen, J. Frey, P. Arm, and M. Hutter, “SMUG Planner: A safe multi-goal planner for mobile robots in challenging environments,” IEEE Robot. Automat. Lett. , vol. 8, no. 11, pp. 7170–7177, 2023

    2023

  16. [24]

    Hybrid map-based path planning for robot navigation in unstructured envi- ronments,

    J. Liu, X. Chen, J. Xiao, S. Lin, Z. Zheng, and H. Lu, “Hybrid map-based path planning for robot navigation in unstructured envi- ronments,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2023, pp. 2216–2223

    2023

  17. [25]

    Traversability-aware adaptive optimization for path planning and control in mountainous terrain,

    S.-W. Yoo, E.-I. Son, and S.-W. Seo, “Traversability-aware adaptive optimization for path planning and control in mountainous terrain,” IEEE Robot. Automat. Lett. , vol. 9, no. 6, pp. 5078–5085, 2024

    2024

  18. [26]

    Where should I walk? predicting terrain properties from images via self-supervised learning,

    L. Wellhausen, A. Dosovitskiy, R. Ranftl, K. Walas, C. Cadena, and M. Hutter, “Where should I walk? predicting terrain properties from images via self-supervised learning,” IEEE Robot. Automat. Lett. , vol. 4, no. 2, pp. 1509–1516, 2019

    2019

  19. [27]

    Safe robot navigation via multi-modal anomaly detection,

    L. Wellhausen, R. Ranftl, and M. Hutter, “Safe robot navigation via multi-modal anomaly detection,” IEEE Robot. Automat. Lett. , vol. 5, no. 2, pp. 1326–1333, 2020

    2020

  20. [28]

    GA-Nav: Efficient terrain segmenta- tion for robot navigation in unstructured outdoor environments,

    T. Guan, D. Kothandaraman, R. Chandra, A. J. Sathyamoorthy, K. Weerakoon, and D. Manocha, “GA-Nav: Efficient terrain segmenta- tion for robot navigation in unstructured outdoor environments,” IEEE Robot. Automat. Lett. , vol. 7, no. 3, pp. 8138–8145, 2022

    2022

  21. [29]

    Path planning with local motion estimations,

    J. Guzzi, R. O. Chavez-Garcia, M. Nava, L. M. Gambardella, and A. Giusti, “Path planning with local motion estimations,” IEEE Robot. Automat. Lett. , vol. 5, no. 2, pp. 2586–2593, 2020

    2020

  22. [30]

    Rough terrain navigation for legged robots using reachability planning and template learning,

    L. Wellhausen and M. Hutter, “Rough terrain navigation for legged robots using reachability planning and template learning,” in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst. , 2021, pp. 6914–6921

    2021

  23. [31]

    ArtPlanner: Robust legged robot navigation in the field,

    ——, “ArtPlanner: Robust legged robot navigation in the field,” J. Field Robot., vol. 3, no. 1, pp. 413–434, 2023

    2023

  24. [32]

    Real-time optimal navigation planning using learned motion costs,

    B. Yang, L. Wellhausen, T. Miki, M. Liu, and M. Hutter, “Real-time optimal navigation planning using learned motion costs,” in Proc. IEEE Int. Conf. Robot. Automat. , 2021, pp. 9283–9289

    2021

  25. [33]

    Prob- abilistic roadmaps for path planning in high-dimensional configuration spaces,

    L. E. Kavraki, P. Svestka, J.-C. Latombe, and M. H. Overmars, “Prob- abilistic roadmaps for path planning in high-dimensional configuration spaces,” IEEE Trans. Robot. , vol. 12, no. 4, pp. 566–580, 1996

    1996

  26. [34]

    A formal basis for the heuristic determination of minimum cost paths,

    P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,” IEEE Trans. Syst. Sci. Cybern. , vol. 4, no. 2, pp. 100–107, 1968

    1968

  27. [35]

    FAST-LIO2: Fast direct LiDAR-inertial odometry,

    W. Xu, Y . Cai, D. He, J. Lin, and F. Zhang, “FAST-LIO2: Fast direct LiDAR-inertial odometry,” IEEE Trans. Robot. , vol. 38, no. 4, pp. 2053–2073, 2022

    2022

  28. [36]

    R. C. Coulter, Implementation of The Pure Pursuit Path Tracking Al- gorithm, Tech. Rep. CMU-RI-TR-92-01, Carnegie Mellon University, 1992

    1992

  29. [37]

    Sampling-based algorithms for optimal motion planning,

    S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,” Int. J. Robot. Res. , vol. 30, no. 7, pp. 846–894, 2011

    2011

  30. [38]

    G. E. Box, G. M. Jenkins, G. C. Reinsel, and G. M. Ljung, Time series analysis: forecasting and control . John Wiley & Sons, 2015

    2015

  31. [39]

    Taking the first step toward autonomous quadruped robots: The Quadruped Robot Challenge at ICRA 2023 in London,

    A. Jacoff, J. Jeon, O. Huke, D. Kanoulas, S. Ha, D. Kim, and H. Moon, “Taking the first step toward autonomous quadruped robots: The Quadruped Robot Challenge at ICRA 2023 in London,” IEEE Robot. Automat. Mag., vol. 30, no. 3, pp. 154–158, 2023

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.