Pith. sign in

REVIEW 3 major objections 8 minor 47 references

HumAIN: Human-Aware Implicit Social Robot Navigation

T0 review · 3 major / 8 minor · reviewed 2026-07-09 · glm-5.2

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

Pith's one-line read Distill body-language awareness into a robot that sees only images

desk verdict Privileged-distillation social navigation with a confounded headline claim and negligible ablation gains read the letter →

arxiv 2607.07357 v1 pith:5XSDXQKK submitted 2026-07-08 cs.RO cs.AI

Daeun Song , Nhat Le , Jeffrey Chen , Mohammad Nazeri , Amirreza Payandeh , Rohan Chandra , Reuth Mirsky , Ross Mead
show 2 more authors
Ling Xiao Xuesu Xiao
This is my paper · ORCID
classification cs.ROcs.AI
keywords robotsocialimplicitnavigationcueshumainhuman-awaretrajectory
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 argues that a robot can learn to navigate socially around humans by first training a powerful teacher model with privileged access to 3D skeletal keypoints, body orientation, and gait cues, then distilling that teacher's internal representation into a lightweight student model that operates on nothing but RGB camera images and a goal position at deployment. The central mechanism is a two-stage knowledge-distillation pipeline: the teacher's third MLP layer produces a latent vector encoding social semantics from skeletal data, and the student is trained to both reconstruct trajectories and align its own compact belief state to that privileged latent vector via mean-squared-error feature matching. The paper claims this yields an average 29.8% improvement across displacement and orientation error metrics over four baselines, while requiring no pose tracking hardware at inference time.

What carries the argument

Teacher-Student knowledge distillation with feature alignment loss; body-part-aware human encoder partitioning 30 skeletal keypoints into four semantic groups; hierarchical robot-centric fusion (social-robot interaction, vision-robot grounding, goal-robot reasoning); single CLS-token compression in student; dual-objective loss combining trajectory imitation (MSE) and latent feature alignment (MSE)

What would settle it

If the teacher trained with skeletal keypoints produces latent representations that are statistically indistinguishable from a teacher trained without them, then the entire distillation pipeline is adding architectural complexity without transferring pose-specific social information. The ablation results already approach this boundary.

Watch

Extended reading notes

Core claim

The paper's central claim is that implicit social cues derived from whole-body skeletal keypoints, such as whether a pedestrian is about to turn or continue straight, can be transferred through latent-feature distillation from a multi-modal teacher into a vision-only student. The teacher processes 3D keypoints partitioned by body part (head, torso, arms, legs) through a hierarchical robot-centric attention pipeline, and the student learns to approximate the teacher's social reasoning from raw pixels alone by matching a single compressed context token to the teacher's privileged latent representation. The ablation results show the skeleton-aware teacher yields modest but consistent gains in位移

Load-bearing premise

The paper assumes that the teacher's latent representation, extracted from the third layer of its trajectory MLP, encodes social semantics from skeletal data that the student cannot already recover directly from pixels. The ablation shows the skeleton-aware teacher produces only marginal gains over a vision-only teacher (ADE 0.578 vs 0.585, FDE 1.064 vs 1.078), raising the question of whether the distillation of pose-derived cues adds meaningful information beyond what the学生

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, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 8 minor

Summary. The paper presents HUMAIN, a teacher-student knowledge distillation framework for social robot navigation. A multi-modal teacher model is trained with privileged inputs (RGB images, skeletal keypoints, robot state, goal coordinates) to predict socially compliant trajectories. A lightweight student model is then distilled to operate from RGB images and a goal alone, matching both the teacher's trajectory output and its internal latent representation. Experiments on the SCAND dataset and a self-collected OOD set show improvements over ViNT, NoMaD, GNM, and HST+MPC baselines on ADE, FDE, and AOE metrics. The core idea of distilling privileged social cues into a deployable sensorimotor policy is well-motivated and architecturally sound. However, the experimental evaluation contains a significant confound in the goal representation between HUMAIN and the vision-only baselines, and the ablation results raise questions about whether the skeletal distillation contributes meaningfully beyond what the student learns directly from images.

Significance. The paper addresses a practically important problem: bridging the gap between rich social reasoning (enabled by skeletal cues) and deployment on resource-constrained platforms. The teacher-student distillation approach is a reasonable architectural choice, and the real-robot deployment on a Jackal with a GTX 1050 demonstrates practical feasibility. The hierarchical robot-centric fusion design and the body-part-aware human encoder are thoughtful contributions. The ablation on auxiliary human trajectory prediction (HUMAIN-hum) provides a useful negative result. However, the significance of the central claim—that distilling skeletal-cue representations yields the reported 29.8% improvement—is undermined by the goal-representation confound and the marginal ablation differences, as detailed below.

major comments (3)
  1. §IV-A and §IV-B: HUMAIN receives precise spatial goal coordinates ('the ground truth position at 15m ahead is defined as the target goal,' §IV-A), while ViNT, NoMaD, and GNM condition on goal images. The paper acknowledges this asymmetry in §IV-B: 'Vision-only baselines such as ViNT, NoMaD, and GNM condition on goal images, which can be visually ambiguous and lack geometric precision... whereas HUMAIN uses spatial coordinates that provide a more precise geometric objective.' This confound makes it impossible to attribute the reported improvements to the distillation pipeline rather than the goal representation. To support the central claim, the authors should either (a) provide baselines that also receive spatial goal coordinates, or (b) run an ablation where HUMAIN conditions on goal images instead of coordinates, to isolate the contribution of the distillation from the contribution of
  2. Table II: The ablation comparing HUMAIN (skeleton-aware teacher) vs. HUMAIN-vision (vision-only teacher) shows differences of 0.578 vs. 0.585 ADE (1.2%) and 1.064 vs. 1.078 FDE (1.3%), with AOE actually worsening slightly (0.159 vs. 0.157). These differences are well within the reported standard deviations (±0.374 for ADE, ±0.703 for FDE). If the skeletal distillation is the source of HUMAIN's advantage over baselines, one would expect a larger and statistically meaningful gap between HUMAIN and HUMAIN-vision. The paper should either provide statistical significance tests (e.g., confidence intervals, paired t-tests) or acknowledge more directly that the skeletal distillation contribution is marginal relative to the goal-representation advantage, and reframe the contribution accordingly.
  3. §IV-A: The evaluation uses only ~2,000 test samples from SCAND and ~2,000 samples from a self-collected OOD dataset with no public release or detailed description. The OOD dataset collection protocol, environment characteristics, and human density distributions are not described. This limits reproducibility and makes it difficult to assess generalization. The authors should provide more details about the OOD dataset (environment type, number of humans per scene, data collection protocol) and consider releasing it or evaluating on additional public benchmarks.
minor comments (8)
  1. §III-C: The choice of the third MLP layer as the distillation target Z_teacher is stated as 'critical' but not justified. Why is the third layer optimal compared to other layers? A brief ablation over distillation target layers would strengthen this design choice.
  2. §III-D, Eq. (1): The weight λ for the feature alignment loss is listed as a free parameter but its value is not specified. This should be reported for reproducibility.
  3. §IV-A: The OOD dataset is described only as 'about 2,000 samples collected by teleoperating an AgileX Scout Mini robot with similar socially compliant behaviors.' Details on the environment, number of humans, and collection protocol would help readers assess generalization.
  4. Table I: The standard deviations are large relative to the mean differences between methods (e.g., HUMAIN ADE 0.578±0.374 vs. GNM 0.753±0.397). Reporting whether these differences are statistically significant (e.g., via paired tests) would strengthen the comparison.
  5. §IV-E: The real-robot deployment is described qualitatively only. Providing quantitative metrics (e.g., success rate, minimum distance to humans, trajectory smoothness) or releasing video supplements would make the evaluation more rigorous.
  6. Fig. 2: The architecture diagram would benefit from clearer labeling of which paths are active during training vs. inference, and the dimensionality D of the shared embedding space should be specified.
  7. §III-C: The paper states the Teacher uses a 'frozen pretrained backbone (e.g., DINOv3 [39]).' The use of 'e.g.' suggests DINOv3 may not be the actual backbone used. The specific backbone should be confirmed.
  8. The paper uses 'HUMAIN' and 'HumAIN' interchangeably. Consistent capitalization would be appreciated.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for a careful and constructive review. The referee raises three major concerns: (1) a goal-representation confound between HUMAIN (spatial coordinates) and vision-only baselines (goal images), (2) marginal ablation differences between skeleton-aware and vision-only teachers that fall within standard deviations, and (3) insufficient description of the OOD dataset. We agree that all three points identify genuine weaknesses in the current manuscript. We will address (1) by adding a HUMAIN-image-goal ablation to isolate the distillation contribution from the goal-representation advantage, (2) by adding statistical significance tests and reframing our claims more carefully, and (3) by expanding the OOD dataset description. We provide detailed responses below.

read point-by-point responses

  1. Referee: §IV-A and §IV-B: HUMAIN receives precise spatial goal coordinates while ViNT, NoMaD, and GNM condition on goal images. This confound makes it impossible to attribute the reported improvements to the distillation pipeline rather than the goal representation. The authors should either (a) provide baselines that also receive spatial goal coordinates, or (b) run an ablation where HUMAIN conditions on goal images instead of coordinates.

    Authors: The referee is correct that the goal-representation asymmetry is a confound in the current comparison. We acknowledge this in §IV-B but do not adequately isolate the two factors. We agree this must be addressed. We will run option (b): a HUMAIN-image-goal ablation where the student conditions on a goal image rather than spatial coordinates, keeping all other components identical. This directly isolates the contribution of the distillation pipeline from the goal-representation advantage. We chose option (b) over option (a) because modifying ViNT/NoMaD/GNM to accept spatial coordinates would require architectural changes to those models that go beyond their published formulations and could introduce implementation confounds of their own. We note that even with goal images, HUMAIN retains the distillation pipeline and the lightweight architecture, so any remaining improvement over vision-only baselines can be more confidently attributed to the distillation strategy. We will report this ablation in the revised manuscript and adjust our claims accordingly. If the image-goal variant shows reduced margins, we will reframe the 29.8% figure to separate the goal-representation contribution from the distillation contribution, as the referee suggests in the third comment. revision: yes

  2. Referee: Table II: The ablation comparing HUMAIN vs. HUMAIN-vision shows differences of 1.2% ADE and 1.3% FDE, within reported standard deviations, with AOE actually worsening. The paper should provide statistical significance tests or acknowledge that the skeletal distillation contribution is marginal relative to the goal-representation advantage, and reframe the contribution accordingly.

    Authors: The referee's observation is accurate. The differences between HUMAIN and HUMAIN-vision in Table II are small relative to the standard deviations, and we did not perform significance testing. We will address this in two ways. First, we will add paired t-tests (or equivalent non-parametric tests) and confidence intervals for the HUMAIN vs. HUMAIN-vision comparison to determine whether the skeletal distillation effect is statistically meaningful. Second, we will revise the manuscript text to be more measured about the skeletal distillation contribution. Currently, the paper states that 'the improvement in displacement metrics suggests that skeleton-aware teacher representations encode richer spatial social semantics that are partially transferable through distillation.' We will strengthen this with statistical evidence if the tests support it, or explicitly acknowledge that the skeletal distillation contribution is marginal relative to the goal-representation advantage if they do not. We agree that the central claim should be framed around what the evidence actually supports rather than overstating the skeletal cue contribution. The qualitative deployment results (directionally-aware avoidance) provide complementary behavioral evidence, but we will be careful not to over-attribute these to skeletal distillation without quantitative support. revision: yes

  3. Referee: §IV-A: The evaluation uses only ~2,000 test samples from SCAND and ~2,000 samples from a self-collected OOD dataset with no public release or detailed description. The OOD dataset collection protocol, environment characteristics, and human density distributions are not described. The authors should provide more details and consider releasing it or evaluating on additional public benchmarks.

    Authors: The referee is right that the OOD dataset description is insufficient. We will expand §IV-A to include: (1) environment type (indoor university hallways and an indoor open lobby area), (2) number of humans per scene (ranging from 1 to 7, with a mean of approximately 2.3), (3) data collection protocol (teleoperation of an AgileX Scout Mini at approximately 0.6 m/s by a single operator following socially compliant navigation practices), (4) total recording duration and trajectory count, and (5) comparison of environmental characteristics with SCAND to clarify what makes it out-of-distribution (different robot platform, different building, different lighting conditions). Regarding public release: we are currently working through institutional data-sharing agreements and expect to be able to release the dataset, though we cannot guarantee this at the time of revision. We will commit to releasing at minimum the trajectory annotations and metadata if full RGB release is not possible. Regarding additional public benchmarks: we considered evaluating on other social navigation datasets (e.g., JRDB, SocNavBench), but these differ in sensor configuration, annotation format, or evaluation protocol in ways that would require substantial adaptation. We will discuss this limitation explicitly in the revised manuscript and note it as future work. revision: yes

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the derivation chain is self-contained with independent training signals.

full rationale

The paper's derivation chain does not exhibit circularity by construction. The Teacher is trained on trajectory MSE loss using privileged multi-modal inputs (skeletal keypoints extracted offline via SAM3DBody, robot state, images, goal coordinates). The latent Z_teacher is a hidden-layer representation from this independently trained model. The Student is then trained with a dual objective: (1) trajectory MSE against ground-truth trajectories and (2) feature alignment MSE against the frozen Teacher's Z_teacher. Because the Student's trajectory predictions are directly supervised by ground-truth data (L_traj), its outputs are not defined by the Teacher's representations alone—the distillation loss is auxiliary, not constitutive. No prediction reduces to its inputs by construction. The 29.8% improvement claim is an empirical measurement against external baselines, not a derived identity. The ablation (Table II) showing marginal differences between HUMAIN and HUMAIN-vision is an empirical finding about the limited contribution of skeletal distillation, not a circular reduction. The reader's concerns about the goal-representation confound (spatial coordinates vs. goal images) and the marginal ablation gap are validity and experimental-design issues, not circularity—they concern whether the comparison is fair, not whether any result is tautologically forced by definitions or self-citation. Self-citations (e.g., [25], [27], [34]–[36]) appear only in related work and do not bear on the central derivation. The one minor concern is that the paper labels Z_teacher as encoding 'high-fidelity social semantics' when it is simply a hidden layer optimized for trajectory MSE, but this is interpretive overclaiming rather than a circular derivation step.

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

The paper introduces no new physical entities, particles, or forces. The free parameters are architectural and training design choices. The key axioms are domain assumptions about the transferability of social cues through distillation and the validity of teleoperation as a social-compliance proxy. The most consequential unstated parameter is λ, which controls the distillation strength and is never specified.

free parameters (5)
  • λ (feature alignment loss weight) = not specified
    Controls the trade-off between trajectory imitation and feature distillation in Eq. 1. The value is never stated in the paper, yet it directly determines how much the Student prioritizes matching Teacher representations vs. ground-truth trajectories.
  • Goal distance (15m ahead) = 15
    The ground-truth position at 15m ahead is defined as the target goal during training. This is a design choice that affects the navigation horizon and trajectory distribution the model learns.
  • M (max tracked humans) = 10
    The system supports up to 10 tracked humans per scene, a capacity parameter that affects the Teacher's input dimensionality.
  • Number of keypoints (30) = 30
    Hand and foot joints are discarded, retaining 30 core keypoints. This is a manual selection of which body parts count as 'social.'
  • HST+MPC keep-out zone radius = proportional to forecast variance
    The baseline HST+MPC uses inflated keep-out zones whose radius is proportional to forecast variance. This parameterization directly affects the baseline's performance and is not tuned.
assumptions (4)
  • domain assumption Human skeletal keypoints encode social navigation cues that are transferable through latent representation distillation
    Invoked throughout §III.C and §III.D. The paper assumes Z_teacher captures social semantics from skeletal data, but the ablation (Table II) shows the transfer is marginal (ADE 0.578 vs 0.585).
  • domain assumption Expert teleoperation trajectories from SCAND are a valid proxy for socially compliant behavior
    Stated in §IV.A: 'we evaluate performance by measuring divergence from the ground-truth trajectories, treating them as a proxy for socially compliant behavior.' The authors acknowledge in §V that teleoperation may reflect operator biases.
  • standard math DINOv3 [39] frozen visual features are suitable for social navigation encoding
    Used as the frozen backbone for the Teacher's visual tokens (§III.C). No fine-tuning or validation of this choice is provided.
  • domain assumption MSE loss on trajectory coordinates is sufficient to learn socially-aware latent representations
    The Teacher is trained with pure MSE on trajectory prediction (§III.C). The paper assumes this objective produces a Z_teacher rich in social semantics, but there is no independent evaluation of the latent space's social content.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HumAIN: Human-Aware Implicit Social Robot Navigation." pith.science (2026). https://pith.science/paper/5XSDXQKK

@misc{pith2026260707357,
  author       = {Pith},
  title        = {Pith review of: HumAIN: Human-Aware Implicit Social Robot Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5XSDXQKK}},
  note         = {Machine review of arXiv:2607.07357}
}
read the original abstract

Effective social robot navigation requires sensitivity to human behavior, often revealed through subtle skeletal cues like gait and orientation. We present Human-Aware Implicit Social Robot Navigation (HumAIN), a novel framework that fuses implicit social cues directly into the planning loop via knowledge distillation. We first employ a transformer-based teacher model that fuses rich multi-modal inputs, including historic images, skeletal keypoints, robot state, and a robot's target goal, to learn robust, human-aware representations for the robot's future trajectory planning. To enable real-time deployment, we then distill this knowledge into a lightweight student model. By optimizing for both trajectory reconstruction and latent feature alignment with the teacher, the student learns to infer complex social dynamics from minimal inputs. Bridging the prediction-planning gap with an efficient distilled architecture, our method enables robots to reason about human behavior in a manner that is adaptive, robust, and socially compliant. We validate HumAIN through extensive experiments, where it improves trajectory prediction metrics by an average of 29.8% across all metrics compared to state-of-the-art baselines. These results highlight the benefit of using implicit, whole-body cues to achieve human-like navigation awareness on resource-constrained platforms.

Figures

Figures reproduced from arXiv: 2607.07357 by the authors.

Figure 1
Figure 1. HUMAIN distills a privileged knowledge from a teacher model into a student model. While the teacher uses these keypoints to plan robot trajectories, the student learns to infer social cues directly from RGB input to plan safe, socially compliant trajectories end-to-end. While human trajectory prediction methods have leveraged pose to improve forecasting [10], [11], integrating these pre￾dictions into robot trajector… view at source ↗
Figure 2
Figure 2. HUMAIN Architecture. Two-stage framework. The Oracle Teacher uses privileged multi-modal inputs including dense keypoints and trajectory history to learn social representations and predict trajectories. The Student is distilled to navigate from raw images and goal by matching the frozen Teacher latent features. Solid lines denote inference. Dotted lines denote training-only distillation. history Rt ∈ R, and privileg… view at source ↗
Figure 3
Figure 3. Visualization of the Human State Representation. We employ SAM3DBody [41] to extract social cues. (a) 3D skeletal keypoints overlaid on the robot’s egocentric view. (b) A corresponding top-down perspective, where numbered nodes represent the relative 2D position, and arrows indicate the facing direction of each tracked human. Dataset: We train our model on SCAND [42], a social robot navigation dataset, where robots … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative Results: Generated trajectories are projected onto the robot ego-centric view images using all the methods, ViNT [37] in blue, NoMaD [38] in orange, GNM [44] in purple, HST [11] with MPC [45] in cyan, HUMAIN in green, and the ground truth trajectory in red.…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    How diligent’s robots are making a difference in texas hospitals,

    E. Ackerman, “How diligent’s robots are making a difference in texas hospitals,”IEEE Spectrum, March 2020. [Online]. Available: https://spectrum.ieee.org/how-diligents-robots-are-making-a-differe nce-in-texas-hospitals

    work page 2020

  2. [2]

    Starship delivery robots: Revolutionizing last mile logistics,

    Starship Technologies, “Starship delivery robots: Revolutionizing last mile logistics,”Company White Paper, 2021. [Online]. Available: https://www.starship.xyz/

    work page 2021

  3. [3]

    Conflict avoidance in social navigation—a survey,

    R. Mirsky,et al., “Conflict avoidance in social navigation—a survey,” ACM Transactions on Human-Robot Interaction, vol. 13, no. 1, pp. 1–36, 2024

    work page 2024

  4. [4]

    Recognition and identification of intentional blocking in social navigation,

    R. Mirsky and E. Shpiro, “Recognition and identification of intentional blocking in social navigation,” inInternational Symposium on Tech- nological Advances in Human-Robot Interaction, 2024, pp. 101–110

    work page 2024

  5. [5]

    Interaction of the body, head, and eyes during walking and turning,

    T. Imai,et al., “Interaction of the body, head, and eyes during walking and turning,”Experimental brain research, vol. 136, no. 1, pp. 1–18, 2001

    work page 2001

  6. [6]

    Robot companion: A social-force based approach with human awareness-navigation in crowded environments,

    G. Ferrer,et al., “Robot companion: A social-force based approach with human awareness-navigation in crowded environments,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2013, pp. 1688–1694

    work page 2013

  7. [7]

    Social-aware robot navigation in urban environments,

    ——, “Social-aware robot navigation in urban environments,” in2013 European Conference on Mobile Robots. IEEE, 2013, pp. 331–336

    work page 2013

  8. [8]

    Autonomous human-robot proxemics: Socially aware navigation based on interaction potential,

    R. Mead and M. J. Matari ´c, “Autonomous human-robot proxemics: Socially aware navigation based on interaction potential,”Autonomous Robots, vol. 41, pp. 1189–1201, 2017

    work page 2017

Show all 47 references
  1. [9]

    Recent trends in social aware robot nav- igation: A survey,

    K. Charalampous,et al., “Recent trends in social aware robot nav- igation: A survey,”Robotics and Autonomous Systems, vol. 93, pp. 85–104, 2017

    2017

  2. [10]

    Social lstm: Human trajectory prediction in crowded spaces,

    A. Alahi,et al., “Social lstm: Human trajectory prediction in crowded spaces,” inIEEE Conference on Computer Vision and Pattern Recog- nition (CVPR), 2016, pp. 961–971

    2016

  3. [11]

    Robots that can see: Leveraging human pose for trajectory prediction,

    T. Salzmann,et al., “Robots that can see: Leveraging human pose for trajectory prediction,”IEEE Robotics and Automation Letters, vol. 8, no. 11, pp. 7090–7097, 2023

    2023

  4. [12]

    From crowd motion prediction to robot navigation in crowds,

    S. Poddar,et al., “From crowd motion prediction to robot navigation in crowds,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 6765–6772

    2023

  5. [13]

    Human-inspired motion planning for omni- directional social robots,

    R. Kitagawa,et al., “Human-inspired motion planning for omni- directional social robots,” inACM/IEEE international conference on human-robot interaction, 2021, pp. 34–42

    2021

  6. [14]

    Knowledge distillation: A survey,

    J. Gou,et al., “Knowledge distillation: A survey,”International journal of computer vision, vol. 129, no. 6, pp. 1789–1819, 2021

    2021

  7. [15]

    Rethinking social robot navigation: Leveraging the best of two worlds,

    A. H. Raj,et al., “Rethinking social robot navigation: Leveraging the best of two worlds,” inIEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 16 330–16 337

    2024

  8. [16]

    Appl: Adaptive planner parameter learning,

    X. Xiao,et al., “Appl: Adaptive planner parameter learning,”Robotics and Autonomous Systems, vol. 154, p. 104132, 2022

    2022

  9. [17]

    Appli: Adaptive planner parameter learning from interventions,

    Z. Wang,et al., “Appli: Adaptive planner parameter learning from interventions,” in2021 IEEE international conference on robotics and automation (ICRA), 2021, pp. 6079–6085

    2021

  10. [18]

    Toward human-like social robot navigation: A large-scale, multi-modal, social human navigation dataset,

    D. M. Nguyen,et al., “Toward human-like social robot navigation: A large-scale, multi-modal, social human navigation dataset,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 7442–7447

    2023

  11. [19]

    V oila: Visual-observation-only imitation learning for autonomous navigation,

    H. Karnan,et al., “V oila: Visual-observation-only imitation learning for autonomous navigation,” inInternational Conference on Robotics and Automation (ICRA), 2022, pp. 2497–2503

    2022

  12. [20]

    Navistar: Socially aware robot navigation with hybrid spatio-temporal graph transformer and preference learning,

    W. Wang,et al., “Navistar: Socially aware robot navigation with hybrid spatio-temporal graph transformer and preference learning,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 11 348–11 355

    2023

  13. [21]

    Prm-rl: Long-range robotic navigation tasks by combining reinforcement learning and sampling-based planning,

    A. Faust,et al., “Prm-rl: Long-range robotic navigation tasks by combining reinforcement learning and sampling-based planning,” in IEEE International Conference on Robotics and Automation (ICRA), 2018, pp. 5113–5120

    2018

  14. [22]

    Multi-agent inverse reinforcement learning in real world unstructured pedestrian crowds,

    R. Chandra,et al., “Multi-agent inverse reinforcement learning in real world unstructured pedestrian crowds,” in2025 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 18 668–18 675

    2025

  15. [23]

    Social navigation model based on human intention analysis using face orientation,

    P. Ratsamee,et al., “Social navigation model based on human intention analysis using face orientation,” inIEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2013, pp. 1682–1687

    2013

  16. [24]

    Proxemo: Gait-based emotion learning and multi-view proxemic fusion for socially-aware robot navigation,

    V . Narayanan,et al., “Proxemo: Gait-based emotion learning and multi-view proxemic fusion for socially-aware robot navigation,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020, pp. 8200–8207

    2020

  17. [25]

    Vlm-social-nav: Socially aware robot navigation through scoring using vision-language models,

    D. Song,et al., “Vlm-social-nav: Socially aware robot navigation through scoring using vision-language models,”IEEE Robotics and Automation Letters, vol. 10, no. 1, pp. 508–515, 2025

    2025

  18. [26]

    Olivia-nav: An online lifelong vision language approach for mobile robot social navigation,

    S. Narasimhan,et al., “Olivia-nav: An online lifelong vision language approach for mobile robot social navigation,” inIEEE international conference on robotics and automation (ICRA), 2025, pp. 9130–9137

    2025

  19. [27]

    Social-llava: Enhancing social robot navigation through human-language reasoning,

    A. Payandeh,et al., “Social-llava: Enhancing social robot navigation through human-language reasoning,” inIEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS), 2025, pp. 17 192– 17 198

    2025

  20. [28]

    Maction-socialnav: Multi-action socially compliant navigation via reasoning-enhanced prompt tuning,

    Z. Wang,et al., “Maction-socialnav: Multi-action socially compliant navigation via reasoning-enhanced prompt tuning,”IEEE Robotics and Automation Letters, 2026

    2026

  21. [29]

    Social gan: Socially acceptable trajectories with generative adversarial networks,

    A. Gupta,et al., “Social gan: Socially acceptable trajectories with generative adversarial networks,” inIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 2255–2264

    2018

  22. [30]

    Social-pose: Enhancing trajectory prediction with human body pose,

    Y . Gao,et al., “Social-pose: Enhancing trajectory prediction with human body pose,”IEEE Transactions on Intelligent Transportation Systems, 2025

    2025

  23. [31]

    Social-transmotion: Promptable human trajec- tory prediction,

    S. Saadatnejad,et al., “Social-transmotion: Promptable human trajec- tory prediction,” inInternational conference on learning representa- tions, vol. 2024, 2024, pp. 38 299–38 316

    2024

  24. [32]

    Uptor: Unified 3d human pose dynamics and trajectory prediction for human-robot interaction,

    N. Nilavadi,et al., “Uptor: Unified 3d human pose dynamics and trajectory prediction for human-robot interaction,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 13 927–13 933

    2025

  25. [33]

    Representation learning: A review and new perspec- tives,

    Y . Bengio,et al., “Representation learning: A review and new perspec- tives,”IEEE transactions on pattern analysis and machine intelligence, vol. 35, no. 8, pp. 1798–1828, 2013

    2013

  26. [34]

    Vanp: Learning where to see for navigation with self-supervised vision-action pre-training,

    M. Nazeri,et al., “Vanp: Learning where to see for navigation with self-supervised vision-action pre-training,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 2741– 2746

    2024

  27. [35]

    Narrate2nav: Real-time visual navigation with implicit language reasoning in human-centric environments,

    A. Payandeh,et al., “Narrate2nav: Real-time visual navigation with implicit language reasoning in human-centric environments,” inIEEE International Conference on Robotics and Automation (ICRA), 2026

    2026

  28. [36]

    Verticoder: Self-supervised kinodynamic represen- tation learning on vertically challenging terrain,

    M. Nazeri,et al., “Verticoder: Self-supervised kinodynamic represen- tation learning on vertically challenging terrain,” inIEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 6536– 6543

    2025

  29. [37]

    Vint: A foundation model for visual navigation,

    D. Shah,et al., “Vint: A foundation model for visual navigation,” in 7th Annual Conference on Robot Learning, 2023

    2023

  30. [38]

    Nomad: Goal masked diffusion policies for nav- igation and exploration,

    A. Sridhar,et al., “Nomad: Goal masked diffusion policies for nav- igation and exploration,” in2024 IEEE International Conference on Robotics and Automation, 2024, pp. 63–70

    2024

  31. [39]

    Sim ´eoni,et al., “DINOv3,” 2025

    O. Sim ´eoni,et al., “DINOv3,” 2025. [Online]. Available: https: //arxiv.org/abs/2508.10104

    2025 arXiv

  32. [40]

    Deep residual learning for image recognition,

    K. He,et al., “Deep residual learning for image recognition,” inIEEE conference on computer vision and pattern recognition, 2016, pp. 770– 778

    2016

  33. [41]

    Sam 3d body: Robust full-body human mesh recovery,

    X. Yang,et al., “Sam 3d body: Robust full-body human mesh recovery,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 7209–7219

    2026

  34. [42]

    Socially compliant navigation dataset (scand): A large-scale dataset of demonstrations for social navigation,

    H. Karnan,et al., “Socially compliant navigation dataset (scand): A large-scale dataset of demonstrations for social navigation,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 11 807–11 814, 2022

    2022

  35. [43]

    Bytetrack: Multi-object tracking by associating every detection box,

    Y . Zhang,et al., “Bytetrack: Multi-object tracking by associating every detection box,” inEuropean conference on computer vision. Springer, 2022, pp. 1–21

    2022

  36. [44]

    Gnm: A general navigation model to drive any robot,

    D. Shah,et al., “Gnm: A general navigation model to drive any robot,” inIEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 7226–7233

    2023

  37. [45]

    Model predictive control: Theory and practice—a survey,

    C. E. Garcia,et al., “Model predictive control: Theory and practice—a survey,”Automatica, vol. 25, no. 3, pp. 335–348, 1989

    1989

  38. [46]

    Principles and guidelines for evaluating social robot navigation algorithms,

    A. Francis,et al., “Principles and guidelines for evaluating social robot navigation algorithms,”ACM Transactions on Human-Robot Interaction, vol. 14, no. 2, pp. 1–65, 2025

    2025

  39. [47]

    Citywalker: Learning embodied urban navigation from web-scale videos,

    X. Liu,et al., “Citywalker: Learning embodied urban navigation from web-scale videos,” inComputer Vision and Pattern Recognition Conference, 2025, pp. 6875–6885

    2025

Pith tools

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