Pith. sign in

REVIEW 4 major objections 6 minor 60 references

OpenGuide: Assistive Object Retrieval in Indoor Spaces for Individuals with Visual Impairments

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read OpenGuide claims that combining a vision-language value map with a POMDP planner nearly doubles multi-object search success over the prior state of the art, and enables recovery from missed detections.

desk verdict Solid systems paper with a promising value-decay idea; the reported gains are plausible but the equations don't match the prose and the real-world validation lacks baselines. read the letter →

arxiv 2509.02425 v1 pith:YCAC5PAG submitted 2025-09-02 cs.RO cs.HC

Yifan Xu , Qianwei Wang , Vineet Kamat , Carol Menassa This is my paper
classification cs.ROcs.HC
keywords assistiveroboticsmulti-objectsearchvisualimpairmentvision-languagemodelsPOMDPplanningfrontierexplorationopen-vocabularynavigation
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

OpenGuide claims that a mobile robot can retrieve several requested objects in unfamiliar indoor spaces by combining a vision-language value map with POMDP-based planning, and that this combination is markedly better than existing frontier-only methods. The paper's core claim is that when the robot misses an object—due to occlusion or a poor viewpoint—it can recover by letting the value map decay and replanning over candidate points, rather than simply moving to the next unexplored frontier. In simulation on a standard 3D-scene benchmark, OpenGuide reports a 55.0% success rate and 0.497 MSPL, versus 28.3% and 0.198 for the strongest baseline, Finder; real-world trials in a home and an office yield a similar 54.2% end-to-end success rate. The practical significance is that open-vocabulary multi-object assistance—such as finding a toothbrush and a sink together—can move from single-object navigation aids to fuller retrieval support for people with visual impairments.

What carries the argument

The carrying mechanism is the multi-layer value map reduced to candidate points: a per-pixel map of VLM similarity, normalized by field-of-view confidence, aggregated across target objects, decayed with a sigmoid function of visit count, then clustered with DBSCAN to produce candidate points. These candidate points simultaneously define the POMDP action space and the discrete belief over each object's location. During planning, POUCT uses a generative model with a distance-based observation function; during real execution, the belief is updated directly from the decayed value map. That asymmetry is what lets OpenGuide 'forget' unproductive high-value areas and revisit potentially missed obje

What would settle it

Measure the calibration of the value map by comparing its ranked candidate points against ground-truth object locations in held-out simulated scenes; if top-ranked candidates contain the target no more often than chance, the decayed-map belief update is steering on noise and the reported success-rate advantage should disappear on new environments.

Watch

Extended reading notes

Core claim

The central claim is that multi-object search in unknown indoor environments can be formulated as a POMDP whose actions and beliefs are built directly from a VLM-generated value map, and that this removes the need for exact Bayesian belief updates during real-world execution. OpenGuide computes per-object cosine-similarity value maps from a pretrained vision-language model, applies a sigmoid decay to suppress repeatedly visited high-value regions, clusters high-value cells into candidate points via DBSCAN, and adds one selected frontier as an exploratory action. POUCT then plans over MoveTo actions in an object-oriented POMDP whose belief is a discrete distribution over candidate points. In

Load-bearing premise

The load-bearing premise is that the vision-language model's similarity values are a sufficiently reliable estimate of where each target object actually is, so that steering by a decayed version of those values can replace a correct probability update during real execution.

Editorial extensions

If this is right

  • Multi-object retrieval becomes an open-vocabulary service: a user can request objects in natural language with no per-scene training or predefined object list.
  • Missed or occluded objects no longer end the search: value decay and replanning lead the robot to re-examine high-value areas it has already passed, where frontier-only methods abandon them.
  • Sequential single-object search is a weak default for multi-object tasks: decomposing a multi-object request into independent VLFM searches yields 12.5% success in simulation versus 55.0% for the joint POMDP approach.
  • POMDP planning is load-bearing: removing it drops success from 55.0% to 27.5%, because the reward balances travel cost, frontier exploration, and target proximity instead of greedily choosing the highest-value point.
  • The mechanism transfers beyond the benchmark: real-world success in a studio apartment and a cluttered office (about 50–57%) tracks the simulated rate, suggesting the approach is not tied to one scene type.

Reading between the lines

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

  • The paper does not test a Bayesian-update baseline with a learned observation model; an implicit consequence is that value-map quality, not planning sophistication, sets the ceiling, and better calibration of VLM similarities should translate directly into search gains.
  • The value-decay update is a heuristic for posterior hedging: it does not distinguish 'this area was searched and the object is absent' from 'this area was searched and the object may have been missed.' A richer belief could condition on detection failures, but that would reintroduce the observation-model dependence the paper avoids.
  • The reported advantage over Finder is concentrated in occlusion recovery; the approach is most likely to be stressed in tasks with many small or visually ambiguous objects, which account for most real-world failures in the paper's own breakdown (11 of 16 perception failures due to detection errors).
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper presents OpenGuide, a mobile robot system for multi-object retrieval in indoor environments to assist people with visual impairments. The system combines a BLIP-2-derived object-value map, Grounding DINO/YOLOv7/SAM object detection, frontier-based exploration, and a POMDP (solved with POUCT) planner. A distinctive component is the replacement of standard Bayesian belief updates during real-world execution by a decayed value map (Eq. 4) that is assumed to encode object-location probabilities. The method is evaluated in Habitat-Matterport3D simulation (120 episodes, 5 scenes) against Random Walk, VLFM, and Finder, with reported success rates of 55.0% vs. Finder's 28.3%; in real-world trials (18 home, 30 office) without baselines; and through ablations on the POMDP and value-decay components.

Significance. If the reported results are robust, OpenGuide would be a strong demonstration of VLM-guided multi-object search with recovery from missed detections, and the real-robot deployment across 48 trials is a useful contribution. However, the central performance claim is currently supported only by a single simulation configuration with no error bars or statistical tests, and the real-world studies lack baselines. The load-bearing belief-update mechanism (Sec. 3.4.1) is formally inconsistent as written and is not validated by calibration evidence. The authors do disclose prior work [23] and state limitations (small/occluded object failures, static VLM grounding) in Sec. 5, which is to their credit.

major comments (4)
  1. [Sec. 3.4.1, Eq. (4) and belief definition] The belief is defined as p_i = v(c_i) / sum_j v'(c_j), mixing the raw value in the numerator with the decayed value in the denominator. This is not normalized and is not a probability distribution over candidates. Moreover, v'(x,y) = 1/(1+exp((u(x,y)-tau)/kappa)) decreases with update count u(x,y), so thresholding v' selects low-update (low-confidence) cells rather than high-object-value cells. As written, candidate extraction and the real-world belief update are internally inconsistent and cannot be reproduced. Since the paper's core claim is that the decayed value map can replace a Bayesian update, this needs to be corrected and the property that p_i forms a valid distribution needs to be restored or justified.
  2. [Sec. 4.2, Table 1] The headline comparison is based on a single configuration: 120 episodes on 5 HM3D scenes, with no error bars, number of seeds, or statistical tests. The claim that OpenGuide 'nearly doubles the success rate and more than doubles the MSPL compared to Finder' therefore has unknown variance; a different scene split or random seed could change the ranking. Please report means and standard deviations over multiple runs, per-scene results, and a significance test (e.g., paired bootstrap or Wilcoxon over episodes) for SR and MSPL.
  3. [Secs. 4.3-4.5] The real-world experiments (18 home trials, 30 office trials) contain no baseline comparisons. The abstract and conclusion claim 'substantial improvements in task success rate and search efficiency over prior methods,' but this is only supported in simulation. The real-world success rates (50.0% and 56.7%) are not compared with VLFM or Finder under matched conditions. Either add a real-world baseline or explicitly restrict the comparative claim to simulation; as written, the claim overreaches.
  4. [Sec. 3.4.1 / Sec. 4.2] The premise that the VLM value map 'already encodes a highly reliable representation of the probability that the object is located at each candidate point' is asserted without calibration evidence. The simulation gains could partly stem from the count-based decay (essentially a novelty bonus) rather than from belief-space reasoning over true object probabilities. The ablation in Table 2 does not isolate this. Provide calibration probes: e.g., correlation between BLIP-2 value and actual object locations on held-out scenes, precision/recall of candidate points, or a comparison of the decayed-map belief against a Bayesian belief computed with a learned observation model. Without this, the mechanism underlying the claimed POMDP advantage remains unsupported.
minor comments (6)
  1. [Sec. 3.4.2, Eq. (5)] The observation function is continuous at d = delta (both branches give 1), which is good, but the parameters delta and beta are not given values or a sensitivity analysis. Please report their settings.
  2. [Sec. 3.3.1] The text states that value maps are 'normalized and aggregated' and that the joint belief is the product of individual beliefs, but no aggregation equation is given. Please specify the exact add/norm operations, since they affect multi-object behavior.
  3. [Sec. 4.2, Eq. (7)] The definition of l_i as 'the optimal shortest path length from the start location to all target objects' is ambiguous in a multi-object setting. Is it the shortest path visiting all target objects in any order, or the sum of individual shortest paths? Please clarify, as MSPL depends on this.
  4. [Sec. 4.3, Table 3] The row with 'fork, bowl, spoon' reports a 0/3 success and average steps 200, which appears to be the maximum step limit. This should be noted explicitly as a cap, and it would be useful to see the breakdown of the three trials.
  5. [Sec. 4.5, Fig. 6] The diagram in Fig. 6 is referenced but not included in the text; please ensure it is available and that the navigation/communication failure categories are defined in the caption.
  6. [Sec. 5 / Data Availability] The limitations paragraph is honest about detector failures, but the Data Availability statement says code is 'available from the corresponding author upon reasonable request.' For a systems paper whose key novelty relies on equations (4)-(6), releasing code or at least hyperparameters would substantially aid reproducibility.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; disclosed OVAMOS self-citation is not load-bearing.

full rationale

OpenGuide's claimed results are benchmarked externally. In simulation, success is defined by actual robot-target distance (<1 m) against ground-truth HM3D object positions (Sec. 4.2), and the baselines are Random Walk, VLFM, and Finder. The value map is produced by a pretrained BLIP-2 model (Sec. 3.3.1), not fitted to the evaluation episodes. The POMDP planner's selected actions are judged by whether the robot physically reaches the target, so the 55.0% vs. 28.3% comparison is not forced by construction. The only self-citation is [23] (OVAMOS), mentioned as 'building upon our previous research in multi-object search' (Sec. 1), and the multi-layer value map add/norm operation (Sec. 3.4) is inherited from prior work. This is transparent provenance, not an unverified uniqueness theorem; the paper's incremental contributions (POMDP planning and value decay) are evaluated via ablation and external comparison. The Sec. 3.4.1 statement that the decayed value map 'already encodes a highly reliable representation' is an empirical design assumption, not a circular definition, and the paper itself discloses perception failures as the dominant bottleneck (Sec. 4.5 and Conclusion). The algebraic inconsistency in Eq. (4) and the normalization following it is a reproducibility/correctness concern, not circularity. Therefore no circular step is identified; the score reflects one non-load-bearing self-citation.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The system rests on domain assumptions about the reliability of VLM value maps, the sufficiency of a distance-based detection model, the static-object assumption, and the claim that a decayed value map can replace a proper Bayesian belief update. Several numeric hyperparameters are introduced without reported values. No new physical entities are postulated.

free parameters (4)
  • Value decay constants tau, kappa = not reported
    Eq. (4) controls how quickly value decays with update count; hand-chosen, no sensitivity analysis.
  • Reward weights lambda_move, lambda_frontier, lambda_target = not reported
    Eq. (6) balances movement cost, frontier exploration, and target proximity; values are not given.
  • Observation model parameters delta, beta = not reported
    Eq. (5) detection probability threshold and decay; not specified, though used in POMDP simulation.
  • Value threshold and DBSCAN cluster parameters = not reported
    Used to extract candidate points from the decayed value map; threshold and eps/min_samples not stated.
assumptions (6)
  • domain assumption The VLM value map encodes a reliable spatial prior for target object locations.
    Invoked in Sections 3.3.1 and 3.4.1; if BLIP-2 similarities do not correlate with true object locations, the planner is navigating to noise.
  • ad hoc to paper The decayed value map can substitute for a Bayesian belief update in real-world execution.
    Stated explicitly in Section 3.4.1 ('there is no longer a need to perform extensive POMDP simulations... the decayed value map directly reflects the updated likelihood'). This is a design assumption without empirical validation.
  • domain assumption The target objects are static and the robot's motion is deterministic within the planning model.
    Transition model in Section 3.4.2 assumes deterministic robot dynamics and static targets; real-world tests show navigation failures (SLAM collisions, getting trapped), so this is an idealization.
  • domain assumption Detections are independent across objects and combine through multiplicative joint belief, as stated in Section 3.3.1, although Section 3.4 says the implementation uses add and norm operations, an unresolved inconsistency.
    Section 3.3.1 follows prior works [44,45,46] and assumes independence across objects when aggregating value maps.
  • domain assumption Objects can be detected by Grounding DINO, YOLOv7, and SAM under open-vocabulary conditions.
    Section 3.3.1 and the real-world results show detection failures are the main bottleneck, so this assumption is only partially satisfied.
  • domain assumption Frontier waypoints extracted from the obstacle map yield efficient exploration.
    Section 3.3.2 and reward function Eq. (6) rely on frontier exploration being useful in novel environments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OpenGuide: Assistive Object Retrieval in Indoor Spaces for Individuals with Visual Impairments." pith.science (2026). https://pith.science/paper/YCAC5PAG

@misc{pith2026250902425,
  author       = {Pith},
  title        = {Pith review of: OpenGuide: Assistive Object Retrieval in Indoor Spaces for Individuals with Visual Impairments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YCAC5PAG}},
  note         = {Machine review of arXiv:2509.02425}
}
read the original abstract

Indoor built environments like homes and offices often present complex and cluttered layouts that pose significant challenges for individuals who are blind or visually impaired, especially when performing tasks that involve locating and gathering multiple objects. While many existing assistive technologies focus on basic navigation or obstacle avoidance, few systems provide scalable and efficient multi-object search capabilities in real-world, partially observable settings. To address this gap, we introduce OpenGuide, an assistive mobile robot system that combines natural language understanding with vision-language foundation models (VLM), frontier-based exploration, and a Partially Observable Markov Decision Process (POMDP) planner. OpenGuide interprets open-vocabulary requests, reasons about object-scene relationships, and adaptively navigates and localizes multiple target items in novel environments. Our approach enables robust recovery from missed detections through value decay and belief-space reasoning, resulting in more effective exploration and object localization. We validate OpenGuide in simulated and real-world experiments, demonstrating substantial improvements in task success rate and search efficiency over prior methods. This work establishes a foundation for scalable, human-centered robotic assistance in assisted living environments.

Figures

Figures reproduced from arXiv: 2509.02425 by the authors.

Figure 1
Figure 1. The OpenGuide system overview 3.2. System Overview The system overview is shown in [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. GuideBot Hardware Design and Demonstration of Usage 4.1. Experiment Hardware Design We developed a GuideBot, a mobile robotic platform designed to assist indi￾viduals, particularly those with visual impairments, in navigating complex indoor environments. The system architecture is illustrated in Fig. 2a. Built on a Turtlebot 2 base, GuideBot uses ROS1 Noetic to enable autonomous mobility, perception, and navigation.… view at source ↗
Figure 3
Figure 3. Qualitative comparison between OpenGuide and Finder [53] in simulation. ability to revisit and refine exploration based on value map information provides OpenGuide with a distinct advantage over Finder in complex search scenarios.We did not compare with VLFM and Random Walk here because their performance is significantly weaker than OpenGuide and Finder. In addition, we conduct an ablation study to analyze the impac… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Qualitative Results of OpenGuide in home environment. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results of OpenGuide in an office environment 20 [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]
Figure 6
Figure 6. Figure 6: Summary of success and failure cases of real world experiment outcomes is visualized in [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 31 canonical work pages

  1. [23]

    Q. Wang, Y. Xu, V. Kamat, C. Menassa, Ovamos: A framework for open- vocabulary multi-object search in unknown environments (2025).arXiv: 2503.02106. URL https://arxiv.org/abs/2503.02106

    arXiv 2025

  2. [1]

    Z. Zou, K. Chen, Z. Shi, Y. Guo, J. Ye, Object detection in 20 years: A survey (2023).arXiv:1905.05055. URL https://arxiv.org/abs/1905.05055

    arXiv 2023

  3. [2]

    URL https://www.cdc.gov/vision-health/data-research/ vision-loss-facts/index.html

    Fast facts: Vision loss. URL https://www.cdc.gov/vision-health/data-research/ vision-loss-facts/index.html

    work page

  4. [3]

    S. E. Chidiac, M. A. Reda, G. E. Marjaba, Accessibility of the built environment for people with sensory disabilities—review quality and representation of evidence, Buildings 14 (3) (2024) 707.doi:10.3390/ buildings14030707. URL http://dx.doi.org/10.3390/buildings14030707

    work page doi:10.3390/buildings14030707 2024

  5. [4]

    Jeamwatthanachai, M

    W. Jeamwatthanachai, M. Wald, G. Wills, Indoor navigation by blind people: Behaviorsandchallengesinunfamiliarspacesandbuildings,British Journal of Visual Impairment 37 (2) (2019) 140–153. doi:10.1177/ 0264619619833723. URL http://dx.doi.org/10.1177/0264619619833723

    work page doi:10.1177/0264619619833723 2019

  6. [5]

    X. Yu, J. Saniie, Visual impairment spatial awareness system for indoor navigation and daily activities, Journal of Imaging 11 (1) (2025) 9.doi: 10.3390/jimaging11010009. URL http://dx.doi.org/10.3390/jimaging11010009

    work page doi:10.3390/jimaging11010009 2025

  7. [6]

    Lahav, D

    O. Lahav, D. Mioduser, Haptic-feedback support for cognitive mapping of unknown spaces by people who are blind, International Journal of Human- ComputerStudies66(1)(2008)23–35. doi:10.1016/j.ijhcs.2007.08. 001. URL http://dx.doi.org/10.1016/j.ijhcs.2007.08.001

    work page doi:10.1016/j.ijhcs.2007.08 2008

  8. [7]

    N. D. Troncoso Aldas, S. Lee, C. Lee, M. B. Rosson, J. M. Carroll, V.Narayanan,Aiguide: Anaugmentedrealityhandguidanceapplicationfor people with visual impairments, in: Proceedings of the 22nd International ACM SIGACCESS Conference on Computers and Accessibility, ASSETS ’20, ACM, 2020, p. 1–13.doi:10.1145/3373625.3417028. URL http://dx.doi.org/10.1145/337...

    work page arXiv 2020

Show all 60 references
  1. [8]

    Valipoor, A

    M. Valipoor, A. de Antonio, J. Cabrera, Analysis and design framework for the development of indoor scene understanding assistive solutions for the person with visual impairment/blindness, Multimedia Systems 30 (3) (May 2024). doi:10.1007/s00530-024-01350-8. URL http://dx.doi....

    2024 doi

  2. [9]

    H. N. Kim, User experience of assistive apps among people with visual impairment, Technology and Disability 34 (3) (2022) 165–174.doi:10. 3233/tad-220377. URL http://dx.doi.org/10.3233/tad-220377

    2022 doi

  3. [10]

    B. Li, J. P. Muñoz, X. Rong, J. Xiao, Y. Tian, A. Arditi, ISANA: Wearable Context-AwareIndoorAssistiveNavigationwithObstacleAvoidanceforthe Blind,SpringerInternationalPublishing,2016,p.448–462. doi:10.1007/ 978-3-319-48881-3_31. URL http://dx.doi.org/10.1007/978-3-319-48881-3_31

    2016 doi

  4. [11]

    D. Sato, U. Oh, K. Naito, H. Takagi, K. Kitani, C. Asakawa, Navcog3: An evaluation of a smartphone-based blind indoor navigation assistant with semantic features in a large-scale environment, in: Proceedings of the 19th International ACM SIGACCESS Conference on Computers and A...

    2017

  5. [12]

    Kaniwa, M

    Y. Kaniwa, M. Kuribayashi, S. Kayukawa, D. Sato, H. Takagi, C. Asakawa, S. Morishima, Chitchatguide: Conversational interaction using large language models for assisting people with visual impairments to explore a shopping mall, Proceedings of the ACM on Human-Computer Interac...

    2024 doi

  6. [13]

    Kayukawa, D

    S. Kayukawa, D. Sato, M. Murata, T. Ishihara, H. Takagi, S. Morishima, C.Asakawa,Enhancingblindvisitor’sautonomyinasciencemuseumusing anautonomousnavigationrobot,in: Proceedingsofthe2023CHIConference on Human Factors in Computing Systems, CHI ’23, ACM, 2023, p. 1–14. doi:10.11...

    2023

  7. [14]

    Chanana, R

    P. Chanana, R. Paul, M. Balakrishnan, P. Rao, Assistive technology solutions for aiding travel of pedestrians with visual impairment, Journal of Rehabilitation and Assistive Technologies Engineering 4 (Jan. 2017). doi:10.1177/2055668317725993. URL http://dx.doi.org/10.1177/205...

    2017 doi

  8. [15]

    1134–1138.doi:10.1145/3610978.3640698

    C.Y.Wong,R.A.Ananto,T.Akiyama,J.P.Nemargut,A.Moon,Perspectives on robotic systems for the visually impaired, in: Companion of the 2024 ACM/IEEEInternationalConferenceonHuman-RobotInteraction,HRI’24, ACM, 2024, p. 1134–1138.doi:10.1145/3610978.3640698. URL http://dx.doi.org/10....

    2024

  9. [16]

    Chen, Y.-Y

    H.-E. Chen, Y.-Y. Lin, C.-H. Chen, I.-F. Wang, Blindnavi: A navigation app forthevisuallyimpairedsmartphoneuser,in: Proceedingsofthe33rdAnnual ACM Conference Extended Abstracts on Human Factors in Computing Systems, CHI ’15, ACM, 2015.doi:10.1145/2702613.2726953. URL http://dx...

    2015

  10. [17]

    doi:10.1145/ 2851613.2851716

    J.-E.Kim,M.Bessho,S.Kobayashi,N.Koshizuka,K.Sakamura,Navigating visually impaired travelers in a large train station using smartphone and bluetooth low energy, in: Proceedings of the 31st Annual ACM Symposium onAppliedComputing,SAC2016,ACM,2016,p.604–611. doi:10.1145/ 2851613....

    2016

  11. [18]

    Murata, D

    M. Murata, D. Ahmetovic, D. Sato, H. Takagi, K. M. Kitani, C. Asakawa, Smartphone-based indoor localization for blind navigation across building complexes,in: 2018IEEEInternationalConferenceonPervasiveComputing and Communications (PerCom), 2018, pp. 1–10.doi:10.1109/PERCOM. 20...

    2018

  12. [19]

    Y. Xu, Z. Luo, Q. Wang, V. Kamat, C. Menassa, Point2graph: An end-to- end point cloud-based 3d open-vocabulary scene graph for robot navigation (2024). arXiv:2409.10350. URL https://arxiv.org/abs/2409.10350

    2024 arXiv

  13. [20]

    doi:10.15607/RSS.2024.XX.077

    A.Werby,C.Huang,M.Büchner,A.Valada,W.Burgard,HierarchicalOpen- Vocabulary3DSceneGraphsforLanguage-GroundedRobotNavigation,in: Proceedings of Robotics: Science and Systems, Delft, Netherlands, 2024. doi:10.15607/RSS.2024.XX.077. 26

    2024 doi

  14. [24]

    Chikhalikar, A

    A. Chikhalikar, A. A. Ravankar, J. Victorio Salazar Luces, Y. Hirata, Semantic-based multi-object search optimization in service robots using probabilisticandcontextualpriors,IEEEAccess12(2024)113151–113164. doi:10.1109/access.2024.3444478. URL http://dx.doi.org/10.1109/ACCESS...

    2024

  15. [25]

    A. Zeng, S. Song, K.-T. Yu, E. Donlon, F. R. Hogan, M. Bauza, D. Ma, O.Taylor,M.Liu,E.Romo,N.Fazeli,F.Alet,N.ChavanDafle,R.Holladay, I. Morona, P. Q. Nair, D. Green, I. Taylor, W. Liu, T. Funkhouser, A. Rodriguez, Robotic pick-and-place of novel objects in clutter with multi- ...

    2019 doi

  16. [26]

    3, IEEE, 2022, p

    A.Shenoy, T.Chen, Y.Sun, Multi-objectgrasping-efficientroboticpicking and transferring policy for batch picking, in: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vol. 3, IEEE, 2022, p. 2741–2747. doi:10.1109/iros47612.2022.9981799. URL http:...

    2022

  17. [27]

    Thiyagarajan, S

    K. Thiyagarajan, S. Kodagoda, M. Luu, T. Duggan-Harper, D. Ritchie, K. Prentice, J. Martin, Intelligent guide robots for people who are blind or have low vision: A review, Vision Rehabilitation International 13 (1) (2022) 1–15. doi:10.2478/vri-2022-0003. URL http://dx.doi.org/...

    2022 doi

  18. [28]

    Fang, Low-cost guide dog robot navigation using dueling dqn, Applied and Computational Engineering 95 (2024) 289–298

    F. Fang, Low-cost guide dog robot navigation using dueling dqn, Applied and Computational Engineering 95 (2024) 289–298. doi:10.54254/ 2755-2721/95/20241751

    2024

  19. [29]

    Galatas, C

    G. Galatas, C. McMurrough, G. L. Mariottini, F. Makedon, eyedog: an assistive-guide robot for the visually impaired, in: Proceedings of the 4th International Conference on PErvasive Technologies Related to Assistive Environments, PETRA ’11, ACM, 2011. doi:10.1145/2141622. 2141...

    2011

  20. [30]

    D. R. Bruno, M. H. de Assis, F. S. Osorio, Development of a mobile robot: Robotic guide dog for aid of visual disabilities in urban environments, in: 2019 Latin American Robotics Symposium (LARS), 2019 Brazilian Symposium on Robotics (SBR) and 2019 Workshop on Robotics in Educ...

    2019

  21. [31]

    R. K. Megalingam, S. Vishnu, V. Sasikumar, S. Sreekumar, Autonomous PathGuidingRobotforVisuallyImpairedPeople,SpringerSingapore,2018, p. 257–266. doi:10.1007/978-981-13-0617-4_25. URL http://dx.doi.org/10.1007/978-981-13-0617-4_25

    2018 doi

  22. [32]

    J. Lai, H. Lei, S. Bao, L. Du, J. Yuan, S. Ma, Design of a portable indoor guide robot for blind people, in: 2021 6th IEEE International Conference on Advanced Robotics and Mechatronics (ICARM), 2021, pp. 592–597. doi:10.1109/ICARM52023.2021.9536077

    2021

  23. [33]

    1–9.doi:10

    S.Kayukawa,T.Ishihara,H.Takagi,S.Morishima,C.Asakawa,Blindpilot: A robotic local navigation system that leads blind people to a landmark object, in: Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, CHI ’20, ACM, 2020, p. 1–9.doi:10. 1145/333...

    2020

  24. [34]

    Slade, A

    P. Slade, A. Tambe, M. J. Kochenderfer, Multimodal sensing and intuitive steeringassistanceimprovenavigationandmobilityforpeoplewithimpaired vision,ScienceRobotics6(59)(Oct.2021). doi:10.1126/scirobotics. 28 abg6594. URL http://dx.doi.org/10.1126/scirobotics.abg6594

    2021 doi

  25. [35]

    S. Cai, A. Ram, Z. Gou, M. A. W. Shaikh, Y.-A. Chen, Y. Wan, K. Hara, S. Zhao, D. Hsu, Navigating real-world challenges: A quadruped robot guiding system for visually impaired people in diverse environments, in: Proceedings of the CHI Conference on Human Factors in Computing S...

    2024

  26. [36]

    iwanttofigure things out

    G.Jain,Y.Teng,D.H.Cho,Y.Xing,M.Aziz,B.A.Smith,"iwanttofigure things out": Supporting exploration in navigation for people with visual impairments, Proc. ACM Hum.-Comput. Interact. 7 (CSCW1) (Apr. 2023). doi:10.1145/3579496. URL https://doi.org/10.1145/3579496

    2023 doi

  27. [37]

    Lacey, S

    G. Lacey, S. MacNamara, Context-aware shared control of a robot mobility aid for the elderly blind, The International Journal of Robotics Research 19 (11) (2000) 1054–1065.doi:10.1177/02783640022067968. URL http://dx.doi.org/10.1177/02783640022067968

    2000 doi

  28. [38]

    Takagi, S

    M.Kuribayashi,T.Ishihara,D.Sato,J.Vongkulbhisal,K.Ram,S.Kayukawa, H. Takagi, S. Morishima, C. Asakawa, Pathfinder: Designing a map-less navigation system for blind people in unfamiliar buildings, in: Proceedings ofthe2023CHIConferenceonHumanFactorsinComputingSystems, CHI ’23, ...

    2023

  29. [39]

    Zheng, Generalized object search (2023).arXiv:2301.10121

    K. Zheng, Generalized object search (2023).arXiv:2301.10121. URL https://arxiv.org/abs/2301.10121

    2023 arXiv

  30. [40]

    S. Wani, S. Patel, U. Jain, A. X. Chang, M. Savva, Multi-on: Benchmarking semanticmapmemoryusingmulti-objectnavigation,in: NeuralInformation Processing Systems (NeurIPS), 2020

    2020

  31. [41]

    Schmalstieg, D

    F. Schmalstieg, D. Honerkamp, T. Welschehold, A. Valada, Learning long- horizon robot exploration strategies for multi-object search in continuous action spaces, in: The International Symposium of Robotics Research, Springer, 2022, pp. 52–66. 29

    2022

  32. [42]

    Sadek, G

    A. Sadek, G. Bono, B. Chidlovskii, A. Baskurt, C. Wolf, Multi-object navigation in real environments using hybrid policies, in: 2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 4085–4091. doi:10.1109/ICRA48891.2023.10161030

    2023

  33. [43]

    Schmalstieg, D

    F. Schmalstieg, D. Honerkamp, T. Welschehold, A. Valada, Learning hierarchical interactive multi-object search for mobile manipulation, IEEE RoboticsandAutomationLetters8(12)(2023)8549–8556. doi:10.1109/ LRA.2023.3329619

    2023

  34. [44]

    Wandzel, Y

    A. Wandzel, Y. Oh, M. Fishman, N. Kumar, L. L. Wong, S. Tellex, Multi-object search using object-oriented pomdps, in: 2019 International Conference on Robotics and Automation (ICRA), 2019, pp. 7194–7200. doi:10.1109/ICRA.2019.8793888

    2019

  35. [45]

    2022–2029.doi: 10.1109/IROS51168.2021.9636737

    K.Zheng,Y.Sung,G.Konidaris,S.Tellex,Multi-resolutionpomdpplanning for multi-object search in 3d, in: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021, pp. 2022–2029.doi: 10.1109/IROS51168.2021.9636737

    2021

  36. [46]

    Zheng, A

    K. Zheng, A. Paul, S. Tellex, A system for generalized 3d multi- object search, in: 2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 1638–1644. doi:10.1109/ICRA48891. 2023.10161387

    2023

  37. [47]

    K. Zhou, K. Zheng, C. Pryor, Y. Shen, H. Jin, L. Getoor, X. E. Wang, Esc: Exploration with soft commonsense constraints for zero-shot object navigation (2023).arXiv:2301.13166. URL https://arxiv.org/abs/2301.13166

    2023 arXiv

  38. [48]

    cat-shaped mug

    V. S. Dorbala, J. F. Mullen, D. Manocha, Can an embodied agent find your “cat-shaped mug”? llm-based zero-shot object navigation, IEEE Robotics andAutomationLetters9(5)(2024)4083–4090. doi:10.1109/LRA.2023. 3346800

    2024 doi

  39. [49]

    B. Yu, H. Kasaei, M. Cao, L3mvn: Leveraging large language models for visual target navigation, in: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 3554–3560.doi:10. 1109/IROS55552.2023.10342512. 30

    2023

  40. [50]

    Majumdar, G

    A. Majumdar, G. Aggarwal, B. Devnani, J. Hoffman, D. Batra, Zson: Zero- shot object-goal navigation using multimodal goal embeddings, in: Neural Information Processing Systems (NeurIPS), 2022

    2022

  41. [51]

    Yokoyama, S

    N. Yokoyama, S. Ha, D. Batra, J. Wang, B. Bucher, Vlfm: Vision-language frontiermapsforzero-shotsemanticnavigation,in: InternationalConference on Robotics and Automation (ICRA), 2024

    2024

  42. [52]

    Kuang, H

    Y. Kuang, H. Lin, M. Jiang, Openfmnav: Towards open-set zero-shot object navigation via vision-language foundation models (2024).arXiv:2402. 10670. URL https://arxiv.org/abs/2402.10670

    2024 arXiv

  43. [53]

    D. Choi, A. Fung, H. Wang, A. H. Tan, Find everything: A general vision language model approach to multi-object search (2024). arXiv: 2410.00388. URL https://arxiv.org/abs/2410.00388

    2024 arXiv

  44. [54]

    Yedidsion, J

    H. Yedidsion, J. Suriadinata, Z. Xu, S. Debruyn, P. Stone, A scavenger hunt for service robots, in: 2021 IEEE International Conference on Robotics and Automation (ICRA), 2021, pp. 7774–7780. doi:10.1109/ICRA48506. 2021.9561722

    2021

  45. [55]

    J. Li, D. Li, S. Savarese, S. Hoi, BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models, in: A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, J. Scarlett (Eds.), Proceedingsofthe40thInternationalConferenceonMachineLea...

    2023

  46. [56]

    S. Liu, Z. Zeng, T. Ren, F. Li, H. Zhang, J. Yang, C. Li, J. Yang, H. Su, J.Zhu,etal.,Groundingdino: Marryingdinowithgroundedpre-trainingfor open-set object detection, arXiv preprint arXiv:2303.05499 (2023)

    2023 arXiv

  47. [57]

    C.-Y. Wang, A. Bochkovskiy, H.-Y. M. Liao, Yolov7: Trainable bag- of-freebies sets new state-of-the-art for real-time object detectors, in: 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 7464–7475.doi:10.1109/CVPR52729.2023.00721. 31

    2023

  48. [58]

    Zhang, D

    C. Zhang, D. Han, Y. Qiao, J. U. Kim, S.-H. Bae, S. Lee, C. S. Hong, Faster segment anything: Towards lightweight sam for mobile applications, arXiv preprint arXiv:2306.14289 (2023)

    2023 arXiv

  49. [59]

    Deng, Dbscan clustering algorithm based on density, in: 2020 7th International Forum on Electrical Engineering and Automation (IFEEA), 2020, pp

    D. Deng, Dbscan clustering algorithm based on density, in: 2020 7th International Forum on Electrical Engineering and Automation (IFEEA), 2020, pp. 949–953.doi:10.1109/IFEEA51475.2020.00199

    2020

  50. [60]

    Silver, J

    D. Silver, J. Veness, Monte-carlo planning in large pomdps, in: J. Lafferty, C. Williams, J. Shawe-Taylor, R. Zemel, A. Culotta (Eds.), Advances in Neural Information Processing Systems, Vol. 23, Curran Associates, Inc., 2010. URL https://proceedings.neurips.cc/paper_files/pap...

    2010

  51. [61]

    S. K. Ramakrishnan, A. Gokaslan, E. Wijmans, O. Maksymets, A. Clegg, J. M. Turner, E. Undersander, W. Galuba, A. Westbury, A. X. Chang, M. Savva, Y. Zhao, D. Batra, Habitat-matterport 3d dataset (HM3d): 1000 large-scale 3d environments for embodied AI, in: Thirty-fifth Conferen...

    2021 arXiv

  52. [62]

    T. Ren, S. Liu, A. Zeng, J. Lin, K. Li, H. Cao, J. Chen, X. Huang, Y. Chen, F. Yan, Z. Zeng, H. Zhang, F. Li, J. Yang, H. Li, Q. Jiang, L. Zhang, Grounded sam: Assembling open-world models for diverse visual tasks (2024). arXiv:2401.14159. 32

    2024 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.