Pith. sign in

REVIEW 2 major objections 56 references

A goal given once at the start is enough for an end-to-end policy to navigate dense clutter and mazes on a quadruped.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-27 12:59 UTC pith:DV4ZUE46

load-bearing objection GUIDE frames a goal-initialized navigation problem for legged robots using proprioceptive spatial anchors, but the abstract supplies zero numbers or comparisons so the claims stay uncheckable. the 2 major comments →

arxiv 2606.10832 v1 pith:DV4ZUE46 submitted 2026-06-09 cs.RO

GUIDE: Goal-Initialized Directional Understanding for End-to-End Visual Navigation

Liang Wang , Jin Jin , KanZhong Yao , YiBin Wu , Fangqiang Ding , Jin Wang , Jun Wu , Zhe Sun
show 1 more author
Qiuguo Zhu
This is my paper
classification cs.RO
keywords end-to-end visual navigationreinforcement learninglegged robotsegomotion estimationspatial memorygoal-initialized navigationquadruped locomotionproprioceptive history
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that legged-robot visual navigation can succeed without continuous goal updates from external state estimators. GUIDE builds an internal directional sense by feeding multi-frequency proprioceptive history into a spatial anchor predictor and raw depth into a local geometry encoder. A reader would care because this removes extra sensors and computation while avoiding myopic traps in partial views. The claim is tested in both simulation and hardware on a quadruped moving through clutter and structured layouts. If correct, robots could operate with only a single initial target and intrinsic memory.

Core claim

GUIDE is a fully end-to-end reinforcement learning framework that cultivates internal directional awareness by incorporating a spatial anchor predictor leveraging multi-frequency proprioceptive history to extract egomotion representations, thereby maintaining a persistent long-horizon spatial context for navigation, while simultaneously using raw depth streams to perceive local environmental geometry.

What carries the argument

The spatial anchor predictor, which turns multi-frequency proprioceptive history into egomotion representations that sustain long-horizon spatial context under partial observability.

Load-bearing premise

Multi-frequency proprioceptive history by itself is enough for the spatial anchor predictor to produce egomotion signals that keep directional awareness intact over long episodes.

What would settle it

A controlled trial in which the robot receives only single-frequency proprioception and then repeatedly loses its way or enters dead-end loops in the same maze layouts would falsify the claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The deployed policy navigates dense clutter and structured mazes without any further goal signals or prior maps.
  • Reliable egomotion and directional awareness emerge from intrinsic spatial memory alone.
  • The same network runs end-to-end in both simulation and on real quadruped hardware.
  • No hierarchical state estimation module is required after the initial goal is supplied.

Where Pith is reading between the lines

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

  • The method may lower the sensor budget for field-deployed legged robots by removing external pose estimators.
  • Similar internal-memory designs could be tested on other platforms whose proprioception shares comparable frequency content.
  • Longer autonomous missions become feasible if the spatial anchor continues to function when depth is intermittently lost.
  • The approach directly addresses partial-observability memory problems that appear in many other sequential decision tasks.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The manuscript proposes GUIDE, a fully end-to-end reinforcement learning framework for goal-initialized visual navigation on legged robots. In this setting the goal is supplied only at episode start; the policy must then rely on intrinsic spatial memory. GUIDE combines a spatial anchor predictor that processes multi-frequency proprioceptive history to extract egomotion representations and maintain long-horizon directional context with raw depth inputs for local geometry. The authors claim that the resulting policy navigates dense clutter and structured mazes without subsequent goal updates or prior maps, and they report evaluation in both simulation and real-world quadruped experiments.

Significance. If the empirical claims are substantiated, the work would demonstrate that proprioceptive history alone can furnish a persistent spatial anchor sufficient for long-horizon navigation, thereby removing the need for continuous external state-estimation modules. This could simplify deployment pipelines and reduce sensory/computational overhead on resource-limited platforms. The absence of any quantitative results in the supplied text, however, prevents assessment of whether the claimed reliability is actually achieved.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'experiments show that GUIDE learns reliable egomotion and directional awareness, enabling a fully end-to-end deployed policy to safely navigate through dense clutter and structured mazes' is unsupported by any quantitative metrics, success rates, baselines, ablation results, or statistical comparisons. Without these data the soundness of the contribution cannot be evaluated.
  2. [Abstract] The weakest assumption identified—that multi-frequency proprioceptive history alone suffices for the spatial anchor predictor to sustain long-horizon context under partial observability—is stated but never tested against alternative proprioceptive encodings or against ground-truth egomotion; no ablation or sensitivity analysis is supplied to support this design choice.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed review and constructive comments on our manuscript. We acknowledge that the abstract claims require stronger quantitative backing in the provided text and that additional ablations would better support the design choices. We will revise the manuscript to address these points directly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'experiments show that GUIDE learns reliable egomotion and directional awareness, enabling a fully end-to-end deployed policy to safely navigate through dense clutter and structured mazes' is unsupported by any quantitative metrics, success rates, baselines, ablation results, or statistical comparisons. Without these data the soundness of the contribution cannot be evaluated.

    Authors: We agree that the abstract's claims must be supported by quantitative evidence for the contribution to be properly evaluated. The full manuscript contains simulation and real-world results with success rates, baseline comparisons, and statistical details; however, since these were not evident in the supplied text, we will revise the abstract to incorporate key metrics (e.g., navigation success rates in clutter and mazes) and ensure the results section is clearly linked to the claims. revision: yes

  2. Referee: [Abstract] The weakest assumption identified—that multi-frequency proprioceptive history alone suffices for the spatial anchor predictor to sustain long-horizon context under partial observability—is stated but never tested against alternative proprioceptive encodings or against ground-truth egomotion; no ablation or sensitivity analysis is supplied to support this design choice.

    Authors: We agree that explicit ablations and sensitivity analyses would strengthen the justification for the multi-frequency proprioceptive encoding. We will add these experiments in the revised manuscript, including comparisons to alternative encodings (e.g., single-frequency or raw IMU) and validation against ground-truth egomotion where available, to directly test the assumption under partial observability. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript is a high-level architectural proposal for an RL navigation policy. No equations, parameter-fitting procedures, uniqueness theorems, or derivation chains appear in the abstract or descriptive sections. Claims about the spatial anchor predictor are presented as empirical outcomes of training rather than reductions to prior fitted quantities or self-citations. The work is therefore self-contained against external benchmarks with no detectable circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract contains no explicit free parameters, axioms, or invented entities; the framework is described conceptually without numerical fitting or new postulated objects.

pith-pipeline@v0.9.1-grok · 5769 in / 1055 out tokens · 27361 ms · 2026-06-27T12:59:50.467760+00:00 · methodology

0 comments
read the original abstract

Learning-based visual navigation for legged robots typically relies on continuous goal updates from hierarchical state estimation to provide a persistent directional reference. This reliance incurs additional sensory and computational overhead and deviates from fully end-to-end mobile autonomy. Furthermore, under partial observability, policies are prone to learn myopic behaviors, easily becoming trapped in dead ends and complex structural layouts. To address these limitations, we investigate a goal-initialized navigation setting, where the target is provided only once at the beginning of an episode, requiring the robot to operate based on intrinsic spatial memory without subsequent goal updates from external modules. In this work, we propose GUIDE, a fully end-to-end reinforcement learning framework designed to cultivate internal directional awareness. Specifically, GUIDE incorporates a spatial anchor predictor that leverages multi-frequency proprioceptive history to extract egomotion representations, thereby maintaining a persistent long-horizon spatial context for navigation. Concurrently, it utilizes raw depth streams to perceive local environmental geometry. We evaluate the proposed framework across both simulation and real-world scenarios on a quadruped robot. Experiments show that GUIDE learns reliable egomotion and directional awareness, enabling a fully end-to-end deployed policy to safely navigate through dense clutter and structured mazes without subsequent goal guidance or prior maps.

Figures

Figures reproduced from arXiv: 2606.10832 by Fangqiang Ding, Jin Jin, Jin Wang, Jun Wu, Kanzhong Yao, Liang Wang, Qiuguo Zhu, YiBin Wu, Zhe Sun.

Figure 1
Figure 1. Figure 1: Our GUIDE framework cultivates internal egomotion and directional awareness for end [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the GUIDE framework. Multi-frequency proprioceptive history is pro￾cessed into proprioceptive tokens, which are supervised to predict spatial anchor vectors to cultivate egomotion and directional awareness. Concurrently, depth buffers are encoded and fused with these tokens via cross-attention to yield spatial latents. Finally, the Actor aggregates these representations with the latest proprioc… view at source ↗
Figure 3
Figure 3. Figure 3: (a1, b1): Cluttered and maze terrains, with curves in gradient color indicating the robot’s trajectories. (a2, b2): Corresponding image inputs for the privileged critic, displaying the two￾channel downsampled maps (Mocc and Mexp) within the same image. 3.3 Reward Design and Multi-Critic Value Estimation Reward Function Design Our navigation reward design combines sparse task signals with dense progress gui… view at source ↗
Figure 4
Figure 4. Figure 4: Quantitative results of real-world en￾vironments. The robot will halt when the inter￾nally estimated distance to the goal is lower than 0.2m. GUIDE achieves the best results across all metrics. Besides, an average FD around 0.2 and 0.3m represents a 0-to-0.1 meter average goal es￾timation error, which further proves the policy’s accuracy and sim-to-real zero-shot generalization ability. In-Lab Evaluations:… view at source ↗
Figure 5
Figure 5. Figure 5: Real-world deployments. GUIDE successfully navigates through (a) in-lab cluttered environments and (b) mazes, as well as unstructured scenarios like (c) long office corridors and outdoor grasslands with (d) dense vegetation and (e) dynamic obstacles. Top-left insets display global maps for (a) and (b) and real-time depth frames for (c-e). In-the-wild Deployments: We further deploy GUIDE in unstructured env… view at source ↗
Figure 6
Figure 6. Figure 6: A representative failure case under lim [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Depth observation processing pipeline. (a) Raw depth image rendered by NVIDIA Warp. (b) Depth image after additive Gaussian noise. (c) Depth Dropout. (d) Quantization. (e) Final policy input after clipping, normalization, cropping, and average-pooling downsampling. Visual Observations. During simulation, the raw depth images are rendered via a lightweight ray￾tracing pipeline built on NVIDIA Warp. Before b… view at source ↗
Figure 8
Figure 8. Figure 8: Training curves for the critic abla￾tion. Compared to the w/o Multi-Critic vari￾ant, GUIDE demonstrates accelerated conver￾gence and achieves a higher total episode reward, highlighting the effectiveness of the Multi-Critic (MuC) architecture. Critic Ablation Analysis. To further investi￾gate the efficacy of the MuC architecture, we compare the training curves of GUIDE with the w/o Multi-Critic variant int… view at source ↗
Figure 9
Figure 9. Figure 9: Terrain curriculum progression. The terrain difficulty progressively increases from left to right for both cluttered and maze environments. These layouts also serve as examples of the eval￾uation benchmarks introduced in Section 4.1. Training Curriculum. Learning to navigate such complex environments from scratch is highly sample-inefficient. Therefore, we em￾ploy a dual-curriculum strategy to guide the op… view at source ↗
Figure 10
Figure 10. Figure 10: Additional real-world maze deployments. The figure illustrates three distinct spawn￾to-goal navigation trials within a 12 m×12 m physical maze. For each trial (rows a–c), the left panels (a1, b1, c1) display the wide view of the environment, while the right panels (a2, b2, c2) provide the corresponding side view. The gradient-colored curves represent the robot’s actual executed trajectories. Starting from… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 6 canonical work pages · 3 internal anchors

  1. [1]

    Hoeller, N

    D. Hoeller, N. Rudin, D. Sako, and M. Hutter. Anymal parkour: Learning agile navigation for quadrupedal robots.Science Robotics, 9(88):eadi7566, 2024

    2024

  2. [2]

    Cheng, K

    X. Cheng, K. Shi, A. Agarwal, and D. Pathak. Extreme parkour with legged robots. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 11443–11450. IEEE, 2024

    2024

  3. [3]

    Y . Wu, J. Kuang, S. Khorshidi, X. Niu, L. Klingbeil, M. Bennewitz, and H. Kuhlmann. Doglegs: Robust proprioceptive state estimation for legged robots using multiple leg-mounted imus. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 8856–8863, 2025

    2025

  4. [4]

    F. Yang, P. Frivik, D. Hoeller, C. Wang, C. Cadena, and M. Hutter. Spatially-enhanced recur- rent memory for long-range mapless navigation via end-to-end reinforcement learning.The International Journal of Robotics Research, page 02783649251401926, 2025

    2025

  5. [5]

    HiPAN: Hierarchical Posture-Adaptive Navigation for Quadruped Robots in Unstructured 3D Environments

    J. Jeong, M. Yoon, S. Choi, H. Shin, T. Yang, and S.-e. Yoon. Hipan: Hierarchical posture- adaptive navigation for quadruped robots in unstructured 3d environments.arXiv preprint arXiv:2604.26504, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  6. [6]

    Vernaza, M

    P. Vernaza, M. Likhachev, S. Bhattacharya, S. Chitta, A. Kushleyev, and D. D. Lee. Search- based planning for a legged robot over rough terrain. In2009 IEEE International Conference on Robotics and Automation, pages 2380–2387. IEEE, 2009

    2009

  7. [7]

    B. Yang, L. Wellhausen, T. Miki, M. Liu, and M. Hutter. Real-time optimal navigation plan- ning using learned motion costs. In2021 IEEE International Conference on Robotics and Automation (ICRA), pages 9283–9289. IEEE, 2021

    2021

  8. [8]

    Gaertner, M

    M. Gaertner, M. Bjelonic, F. Farshidian, and M. Hutter. Collision-free mpc for legged robots in static and dynamic scenes. In2021 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 8266–8272. IEEE, 2021

    2021

  9. [9]

    C. Cao, H. Zhu, F. Yang, Y . Xia, H. Choset, J. Oh, and J. Zhang. Autonomous exploration development environment and the planning algorithms. In2022 International Conference on Robotics and Automation (ICRA), pages 8921–8928. IEEE, 2022

    2022

  10. [10]

    Jiang, H

    P. Jiang, H. Liu, J. Jin, W. Wang, and X. Li. Learning to evolve: Multi-modal interactive fields for robust humanoid navigation in dynamic environments. InRobotics: Science and Systems, 2026

    2026

  11. [11]

    Pfeiffer, M

    M. Pfeiffer, M. Schaeuble, J. I. Nieto, R. Y . Siegwart, and C. Cadena. From perception to de- cision: A data-driven approach to end-to-end motion planning for autonomous ground robots. 2017 IEEE International Conference on Robotics and Automation (ICRA), pages 1527–1533, 2016

    2017

  12. [12]

    Loquercio, E

    A. Loquercio, E. Kaufmann, R. Ranftl, M. M ¨uller, V . Koltun, and D. Scaramuzza. Learning high-speed flight in the wild.Science Robotics, 6(59):eabg5810, 2021

    2021

  13. [13]

    L. Tai, G. Paolo, and M. Liu. Virtual-to-real deep reinforcement learning: Continuous con- trol of mobile robots for mapless navigation. In2017 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 31–36. IEEE, 2017. 9

    2017

  14. [14]

    Wijmans, A

    E. Wijmans, A. Kadian, A. S. Morcos, S. Lee, I. Essa, D. Parikh, M. Savva, and D. Batra. Dd-ppo: Learning near-perfect pointgoal navigators from 2.5 billion frames. InInternational Conference on Learning Representations, 2019

    2019

  15. [15]

    Zhu and M

    W. Zhu and M. Hayashibe. A hierarchical deep reinforcement learning framework with high efficiency and generalization for fast and safe navigation.IEEE Transactions on industrial Electronics, 70(5):4962–4971, 2022

    2022

  16. [16]

    Y . Zhu, R. Mottaghi, E. Kolve, J. J. Lim, A. Gupta, L. Fei-Fei, and A. Farhadi. Target-driven visual navigation in indoor scenes using deep reinforcement learning. In2017 IEEE interna- tional conference on robotics and automation (ICRA), pages 3357–3364. IEEE, 2017

    2017

  17. [17]

    H. Shi, L. Shi, M. Xu, and K.-S. Hwang. End-to-end navigation strategy with deep rein- forcement learning for mobile robots.IEEE Transactions on Industrial Informatics, 16(4): 2393–2402, 2019

    2019

  18. [18]

    Huang, Y

    W. Huang, Y . Zhou, X. He, and C. Lv. Goal-guided transformer-enabled reinforcement learn- ing for efficient autonomous navigation.IEEE Transactions on Intelligent Transportation Sys- tems, 25(2):1832–1845, 2023

    2023

  19. [19]

    W ¨ohlke, F

    J. W ¨ohlke, F. Schmitt, and H. van Hoof. Hierarchies of planning and reinforcement learning for robot navigation. In2021 IEEE international conference on robotics and automation (ICRA), pages 10682–10688. IEEE, 2021

    2021

  20. [20]

    Y . Gao, J. Wu, X. Yang, and Z. Ji. Efficient hierarchical reinforcement learning for mapless navigation with predictive neighbouring space scoring.IEEE Transactions on Automation Science and Engineering, 21(4):5457–5472, 2023

    2023

  21. [21]

    J. Gao, X. Pang, Q. Liu, and Y . Li. Hierarchical reinforcement learning for safe mapless navigation with congestion estimation. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 8849–8855. IEEE, 2025

    2025

  22. [22]

    Pfeiffer, S

    M. Pfeiffer, S. Shukla, M. Turchetta, C. Cadena, A. Krause, R. Siegwart, and J. Nieto. Re- inforced imitation: Sample efficient deep reinforcement learning for mapless navigation by leveraging prior demonstrations.IEEE Robotics and Automation Letters, 3(4):4423–4430, 2018

    2018

  23. [23]

    J. Ye, D. Batra, A. Das, and E. Wijmans. Auxiliary tasks and exploration enable objectgoal navigation. InProceedings of the IEEE/CVF international conference on computer vision, pages 16117–16126, 2021

    2021

  24. [24]

    J. Lee, M. Bjelonic, A. Reske, L. Wellhausen, T. Miki, and M. Hutter. Learning robust au- tonomous navigation and locomotion for wheeled-legged robots.Science Robotics, 9(89): eadi9641, 2024

    2024

  25. [25]

    Q. Yuan, Z. Cao, M. Cao, and K. Li. Reasan: Learning reactive safe navigation for legged robots.arXiv preprint arXiv:2512.09537, 2025

    work page arXiv 2025

  26. [26]

    Sadat, S

    A. Sadat, S. Casas, M. Ren, X. Wu, P. Dhawan, and R. Urtasun. Perceive, predict, and plan: Safe motion planning through interpretable semantic representations. InEuropean Conference on Computer Vision, pages 414–430. Springer, 2020

    2020

  27. [27]

    D. Shah, A. Sridhar, A. Bhorkar, N. Hirose, and S. Levine. Gnm: A general navigation model to drive any robot. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 7226–7233. IEEE, 2023

    2023

  28. [28]

    D. Shah, A. K. Sridhar, N. Dashora, K. Stachowicz, K. Black, N. Hirose, and S. Levine. Vint: A foundation model for visual navigation. InConference on Robot Learning, 2023. 10

    2023

  29. [29]

    F. Yang, C. Wang, C. Cadena, and M. Hutter. iplanner: Imperative path planning. InRobotics: Science and Systems, 2023

    2023

  30. [30]

    P. Roth, J. Nubert, F. Yang, M. Mittal, and M. Hutter. Viplanner: Visual semantic impera- tive learning for local navigation. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 5243–5249. IEEE, 2024

    2024

  31. [31]

    Hoeller, L

    D. Hoeller, L. Wellhausen, F. Farshidian, and M. Hutter. Learning a state representation and navigation in cluttered and dynamic environments.IEEE Robotics and Automation Letters, 6 (3):5081–5088, 2021

    2021

  32. [32]

    Zhang, J

    C. Zhang, J. Jin, J. Frey, N. Rudin, M. Mattamala, C. Cadena, and M. Hutter. Resilient legged local navigation: Learning to traverse with compromised perception end-to-end. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 34–41. IEEE, 2024

    2024

  33. [33]

    T. He, C. Zhang, W. Xiao, G. He, C. Liu, and G. Shi. Agile but safe: Learning collision-free high-speed legged locomotion. InRobotics: Science and Systems, 2024

    2024

  34. [34]

    P. Roth, J. Frey, C. Cadena, and M. Hutter. Learned perceptive forward dynamics model for safe and platform-aware robotic navigation. InRobotics: Science and Systems, 2025

    2025

  35. [35]

    S. Chen, M. Yang, H. Mao, J. Zhang, H. Liu, S. He, D. Zhang, Z. Qiu, and C. Zhang. Sea-nav: Efficient policy learning for safe and agile quadruped navigation in cluttered environments, 2026

    2026

  36. [36]

    Truong, D

    J. Truong, D. Yarats, T. Li, F. Meier, S. Chernova, D. Batra, and A. Rai. Learning naviga- tion skills for legged robots with learned robot embeddings. In2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 484–491. IEEE, 2021

    2021

  37. [37]

    T. Miki, J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter. Learning robust percep- tive locomotion for quadrupedal robots in the wild.Science robotics, 7(62):eabk2822, 2022

    2022

  38. [38]

    P. Long, W. Liu, and J. Pan. Deep-learned collision avoidance policy for distributed multiagent navigation.IEEE Robotics and Automation Letters, 2(2):656–663, 2017

    2017

  39. [39]

    P. Long, T. Fan, X. Liao, W. Liu, H. Zhang, and J. Pan. Towards optimally decentralized multi-robot collision avoidance via deep reinforcement learning. In2018 IEEE international conference on robotics and automation (ICRA), pages 6252–6259. IEEE, 2018

    2018

  40. [40]

    Andrychowicz, F

    M. Andrychowicz, F. Wolski, A. Ray, J. Schneider, R. Fong, P. Welinder, B. McGrew, J. To- bin, O. Pieter Abbeel, and W. Zaremba. Hindsight experience replay.Advances in neural information processing systems, 30, 2017

    2017

  41. [41]

    W. Ding, S. Li, H. Qian, and Y . Chen. Hierarchical reinforcement learning framework towards multi-agent navigation. In2018 IEEE international conference on robotics and biomimetics (ROBIO), pages 237–242. IEEE, 2018

    2018

  42. [42]

    Wijmans, M

    E. Wijmans, M. Savva, I. Essa, S. Lee, A. S. Morcos, and D. Batra. Emergence of maps in the memories of blind navigation agents.AI Matters, 9(2):8–14, 2023

    2023

  43. [43]

    F. Xia, A. R. Zamir, Z. He, A. Sax, J. Malik, and S. Savarese. Gibson env: Real-world per- ception for embodied agents. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 9068–9079, 2018

    2018

  44. [44]

    A. X. Chang, A. Dai, T. A. Funkhouser, M. Halber, M. Nießner, M. Savva, S. Song, A. Zeng, and Y . Zhang. Matterport3d: Learning from rgb-d data in indoor environments.2017 Interna- tional Conference on 3D Vision (3DV), pages 667–676, 2017

    2017

  45. [45]

    E. C. Tolman. Cognitive maps in rats and men.Psychological review, 55(4):189, 1948. 11

    1948

  46. [46]

    O’keefe and L

    J. O’keefe and L. Nadel. Pr ´ecis of o’keefe & nadel’s the hippocampus as a cognitive map. Behavioral and Brain Sciences, 2(4):487–494, 1979

    1979

  47. [47]

    M. Liu, M. Zhu, and W. Zhang. Goal-conditioned reinforcement learning: Problems and solutions.arXiv preprint arXiv:2201.08299, 2022

    work page Pith review arXiv 2022

  48. [48]

    Proximal Policy Optimization Algorithms

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  49. [49]

    R. Avni, Y . Tzvaigrach, and D. Eilam. Exploration and navigation in the blind mole rat (spalax ehrenbergi): global calibration as a primer of spatial representation.Journal of Experimental Biology, 211(17):2817–2826, 2008

    2008

  50. [50]

    R. A. Epstein, E. Z. Patai, J. B. Julian, and H. J. Spiers. The cognitive map in humans: spatial navigation and beyond.Nature neuroscience, 20(11):1504–1513, 2017

    2017

  51. [51]

    J. Long, Z. Wang, Q. Li, L. Cao, J. Gao, and J. Pang. Hybrid internal model: Learning agile legged locomotion with simulated robot response. InInternational Conference on Learning Representations, volume 2024, pages 14084–14100, 2024

    2024

  52. [52]

    Mysore, G

    S. Mysore, G. Cheng, Y . Zhao, K. Saenko, and M. Wu. Multi-critic actor learning: Teaching rl policies to act with style. InInternational Conference on Learning Representations, 2022

    2022

  53. [53]

    L. Wang, K. Yao, Y . Liu, W. Qin, J. Wu, Z. Sun, and Q. Zhu. Puma: Perception-driven unified foothold prior for mobility augmented quadruped parkour.arXiv preprint arXiv:2601.15995, 2026

    work page arXiv 2026

  54. [54]

    Zargarbashi, J

    F. Zargarbashi, J. Cheng, D. Kang, R. Sumner, and S. Coros. Robotkeyframing: Learning locomotion with high-level objectives via mixture of dense and sparse rewards. InConference on Robot Learning, pages 916–932. PMLR, 2025

    2025

  55. [55]

    High-Dimensional Continuous Control Using Generalized Advantage Estimation

    J. Schulman, P. Moritz, S. Levine, M. I. Jordan, and P. Abbeel. High-dimensional continuous control using generalized advantage estimation.CoRR, abs/1506.02438, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  56. [56]

    R. Tarjan. Depth-first search and linear graph algorithms.SIAM Journal on Computing, 1(2): 146–160, 1972. 12 Appendix TABLE OF CONTENTS A Observation Details B Reward Functions C Critic Details D Training Details E Additional Real-World Deployments A Observation Details Proprioceptive Observations.The raw proprioceptive observation frame at time stept, de...

    1972