Socially-Aware Autonomous Doorway Traversal and Payload Delivery for Emergency Assistance

Ananya Trivedi; Andrew Snowdy; Lorena Maria Genua; Sarvesh Prajapati; Taskin Padir

REVIEW 3 major objections 6 minor 14 references

Socially-Aware Autonomous Doorway Traversal and Payload Delivery for Emergency Assistance

T0 review · 3 major / 6 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read A behavior-tree robot can clear doorways and deliver rescue gear for evacuations, finishing 97 of 105 trials that exercise the critical social decisions.

desk verdict Solid systems integration of known pieces into a socially-aware doorway-and-payload stack on the HSR; 97/105 is honestly reported and perception is correctly named as the limiter. read the letter →

arxiv 2607.05315 v2 pith:GJHU6FT5 submitted 2026-07-06 cs.RO

Andrew Snowdy , Ananya Trivedi , Sarvesh Prajapati , Lorena Maria Genua , Taskin Padir This is my paper · ORCID

classification cs.RO

keywords behaviortreessocialnavigationdoortraversalmobilemanipulationemergencyroboticshuman-robotinteractionpayloaddeliveryADAbuttons

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that an indoor service robot can assist emergency evacuation by opening ADA doors ahead of people and by fetching rescue equipment for responders. The authors build a modular system whose core is a behavior tree: perception reports buttons, doors, people, and payloads, the tree selects the next high-level task, and a motion layer executes base and arm commands. The robot must decide, at a doorway, whether to press a button, yield to a nearby person, walk through a door someone else is holding, or first grasp a bottle or flashlight and then traverse. Across 105 hardware and simulation trials that capture those decisions, the system completed 97 successfully, with failures concentrated in perception rather than planning or control. The result is offered as a practical, extensible basis for robots that keep exit paths clear and support the people carrying out a rescue.

What carries the argument

A behavior tree: a hierarchy of selector and sequence nodes that re-evaluate priorities every tick and can preempt a lower-priority action the moment a higher-priority condition becomes true. It turns continuous perception of buttons, doors, people, and payloads into the concrete choices the robot must make at a doorway without enumerating every transition in advance.

What would settle it

Repeat the assisted-traversal trials with a person holding the door at intermediate angles under varied lighting; if the robot still misclassifies the door and wrongly presses the button or fails to pass in a large share of those trials, the binary door-state premise fails.

Watch

Extended reading notes

Core claim

A modular behavior-tree architecture that fuses dual YOLO detectors and LiDAR costmaps with reactive task selection and damped-least-squares arm control can integrate socially-aware doorway traversal and payload delivery well enough to succeed in 97 of 105 trials that exercise the key evacuation decisions: press button, yield, pass through a held door, or retrieve equipment first.

Load-bearing premise

The system treats a door as simply open or closed by checking whether LiDAR sees a flat obstacle inside the camera's door slice, and assumes a small fine-tuned detector generalizes to intermediate angles and real lighting.

Editorial extensions

If this is right

Robots can keep ADA doorways open so evacuating people do not form queues at slow automatic doors.
The same stack can retrieve water bottles or flashlights and deliver them to responders while the building is being cleared.
New emergency behaviors can be added as subtrees without rewriting existing nodes.
Active manipulation can be preempted in under 200 ms when a person enters the workspace, matching human walking timescales at doorways.
Perception quality, not motion planning, is the binding constraint on field deployment.

Reading between the lines

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

Replacing the binary open/closed door model with continuous door-angle estimation would close the main failure mode seen when doors are held only partially open.
Proximity-only yielding can still conflict with people walking parallel to the door; adding short-horizon velocity and heading would tighten social safety.
The same reactive tree could compose multi-room retrieve-and-return chains once localization drift is corrected by loop closure or landmarks.
Expanding the button-and-door training set across finishes and lighting is the highest-leverage next experiment for real buildings.

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.

Desk Editor's Note

read the letter

This is a clean systems paper, not a new algorithm paper. What is new is the integrated behavior-tree executive that jointly handles ADA-button press, human preemption, held-door traversal, and pre-traversal payload grasp, plus the 105-trial characterization on the Toyota HSR with code released. The individual components (BT, YOLO, DLS IK, quintic splines, social doorway negotiation) are already in the literature they cite; the value is the modular coupling and the empirical record of the decisions that matter in an evacuation setting.

They do the evaluation properly. Success rates are broken out by scenario, failures are attributed (almost all perception, not tree logic or IK), timing numbers are measured, and the binary door-state model is openly flagged as the weak point when a human holds the door only partially open. That matches the stress-test note: the load-bearing premise is already diagnosed by the authors themselves in §V.A. No circularity, no over-claim on the 97/105 figure, and the free parameters (damping, velocities, smoothing α, proximity zones) are ordinary engineering knobs, not hidden fitting.

Soft spots are real but secondary. Per-scenario N is modest (mostly 10–15), there are no algorithmic baselines, the abstract says five hardware scenarios while the body says four, and the 600-image fine-tune plus binary open/closed classification will not travel far outside their building. None of that invalidates the reported results; it just keeps the paper in the “useful engineering baseline” category rather than a foundational result.

Who it is for: people building indoor service or emergency-assist robots who need a concrete, extensible BT reference with measured preemption latency and failure modes. It deserves a serious referee. I would accept for peer review, expect the usual requests for more baselines and continuous door-angle estimation, and would bring it to reading group if we are talking mobile manipulation stacks. I would cite the empirical numbers and the architecture if I were writing a related systems paper in the next year.

Referee Report

3 major / 6 minor

Summary. The paper presents a modular Behavior Tree architecture for socially-aware autonomous doorway traversal and payload delivery on the Toyota HSR, motivated by robot-assisted emergency evacuation. Perception (LiDAR costmap + dual YOLOv11n streams for buttons/doors/people/payloads, with glass-door LiDAR fusion), a reactive BT executive (Selectors for preemption, button type, door state, person intent), and a motion layer (DLS IK with warm-start, quintic splines, feedforward-PD) are integrated via a shared Blackboard. Evaluation comprises 105 trials across four hardware scenarios (S1 nominal, S2 human obstruction, S3 assisted/held door, S4a/b pre-traversal grasp of bottle/flashlight) and three simulation scenarios (S5 full multi-door retrieve-and-return chain, S6 preemption, S7 obstacle + assisted entry). The system succeeds in 97/105 trials; failures are attributed almost exclusively to perception (binary door-state estimation under partial openings; object detection under occlusion/lighting). Timing (perception <200 ms, IK ~20 ms warm-started, BT 100 ms, preemption <200 ms) is reported, and code/videos are released.

Significance. If the empirical claim holds under the stated conditions, the work supplies a concrete, reproducible systems baseline that unifies door-button manipulation, payload grasp, and human-reactive yielding at doorways—capabilities previously treated in isolation. The BT’s demonstrated preemption latency and compositional multi-door chain (S5) are useful engineering results for service-robot evacuation assistance. Explicit failure attribution to perception rather than planning, measured subsystem latencies, warm-started IK, and public code strengthen the contribution relative to typical systems papers that report only aggregate success. The result is incremental rather than foundational, but it is a practical, falsifiable step toward deployable emergency-assistance robots.

major comments (3)

Abstract vs. body inconsistency on scenario count: the abstract states “five hardware and three simulation scenarios,” while §IV and Table II describe four hardware scenarios (S1–S4, with S4 split into 4a/4b) plus three simulation scenarios. This is a load-bearing presentation error for the central 105-trial claim and must be corrected for reproducibility.
§IV.A.3 (S3) and §V.A: the binary open/closed door-state classifier (LiDAR planarity inside the YOLO angular slice) is the dominant failure mode (2/15 failures when the door is only partially held open). The paper correctly flags this and proposes continuous angle estimation as future work, yet the assisted-traversal and full-chain claims rest on this premise. A quantitative characterization of intermediate-angle performance (or an ablation replacing the binary classifier) is needed before the “practical basis” claim can be fully accepted for realistic held-door conditions.
Experimental design (§IV, Table II): per-scenario N is modest (10–15 hardware trials) and no algorithmic baselines (e.g., FSM executive, alternative door-state estimators, or prior door-traversal planners) are reported. While the absolute success rates and failure attribution are useful, the absence of comparative numbers limits assessment of whether the BT architecture itself, rather than the perception/motion stack, drives the observed performance.

minor comments (6)

§I and abstract: “four hardware” / “five hardware” mismatch already noted; also ensure the abstract’s “105 trials … five hardware” is aligned with Table II’s 15+15+15+30+10+10+10 = 105.
Fig. 4 caption and tree diagram: node types are defined in a legend, but the Reactive check / IsPathClear condition that implements the <200 ms preemption path is not labeled on the figure itself; a call-out would help readers map the orange arrow of Fig. 2 to the tree.
Eq. (1): the damping factor λ = 0.5 and step α = 0.1 are stated without units or sensitivity discussion; a brief note on how they were chosen (or that they are standard DLS defaults) would aid reproduction.
Related Work: Thomas et al. [6] is cited for doorway negotiation, yet the bibliography entry appears as “Messing et al.”; verify author list consistency.
Table I: “ADA Automatic Door Timeout Window ~10.0 s” is listed as a measured system characteristic; clarify that this is an environmental constraint rather than a measured subsystem latency.
§V.C: the single S5 failure is attributed to “localization drift,” but no quantitative drift metric or map-update rate is given; a short sentence would strengthen the claim that the BT executive itself is sound.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical systems evaluation whose 97/105 success claim is measured against external trial outcomes, not forced by definition or self-citation.

full rationale

This paper is a modular robotics systems evaluation (perception + BehaviorTree.CPP executive + DLS IK / quintic-spline motion) validated by 105 hardware and simulation trials. The central claim is an observed success count (97/105) against externally defined scenario outcomes (press button, yield, pass held door, retrieve payload). There is no mathematical derivation in which a quantity is defined from the same data it is said to predict; no fitted parameter is renamed a prediction; and no uniqueness theorem or ansatz is imported from overlapping authors to force the result. The sole author-overlapping citation ([7], Trivedi et al. on chance-constrained MPC for quadrupeds) appears only as related-work background on a different platform and does not enter the door-traversal logic, the success metric, or any load-bearing premise. Perception failures (binary door state, YOLO fine-tune limits) are reported as empirical limitations, not hidden by circular construction. The derivation chain is therefore self-contained against external benchmarks; circularity score is 0.

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

As a systems paper the central claim rests on engineering choices and domain assumptions rather than free theoretical parameters or invented physical entities. The free parameters are controller and perception thresholds chosen by hand or by experiment; the axioms are standard robotics modeling assumptions plus the binary door-state model that the discussion itself flags as brittle.

free parameters (4)

DLS damping λ and step α
Set to λ=0.5 and α=0.1 with joint clamp 0.1 rad; chosen for numerical stability near singularities, not derived from first principles.
Base target velocities and T_min
v_target = 0.2 m/s nominal, 0.4 m/s through door; T_min = 1.5 s. Hand-tuned to clear the ~10 s ADA auto-close window while remaining gentle near people.
Person-tracking exponential smoothing α and prediction horizon
α=0.3 on position/velocity; four-step 0.2 s extrapolation. Heuristic values that determine when IsPathClear fails.
YOLO confidence thresholds and proximity-zone geometry
Unspecified numeric thresholds plus a ground-plane proximity zone that triggers preemption; directly control false-positive rate and social reactivity.

assumptions (4)

ad hoc to paper Door state is adequately captured by a binary open/closed classifier from LiDAR planarity inside the YOLO angular slice.
Stated in the perception module and used by the BT; the discussion explicitly notes failure on intermediate angles (S3).
domain assumption ADA automatic doors remain open for approximately 10 s after button press.
Used to set the higher traversal velocity; treated as a fixed environmental constant.
domain assumption A fine-tuned YOLOv11n on 600 images plus a general YOLO stream supply sufficiently accurate detections for real-time BT decisions at ~8 Hz / <200 ms latency.
Core perception premise; failures in S3 and S4 are attributed to its violation under partial occlusion or intermediate door pose.
domain assumption Human intent at a doorway can be approximated by spatial proximity plus short-horizon velocity smoothing for preemption decisions.
Implements the social-yield behavior; no full intent model is claimed.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Socially-Aware Autonomous Doorway Traversal and Payload Delivery for Emergency Assistance." pith.science (2026). https://pith.science/paper/GJHU6FT5

@misc{pith2026260705315,
  author       = {Pith},
  title        = {Pith review of: Socially-Aware Autonomous Doorway Traversal and Payload Delivery for Emergency Assistance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GJHU6FT5}},
  note         = {Machine review of arXiv:2607.05315}
}

read the original abstract

In this work, we focus on the scenario of a robot-assisted emergency evacuation. We consider two capabilities relevant to such a setting. The first is opening doors ahead of the people being evacuated, so that their path toward an exit stays clear. The second is retrieving rescue equipment and delivering it to the emergency responders carrying out the evacuation. From a systems perspective, this involves several tasks at once. The robot must locate ADA-compliant door buttons and the rescue equipment it needs to retrieve. Additionally, it must remain aware of the people around it and adapt its behavior to them, so that it supports the evacuation rather than getting in the way. We address these demands with a behavior tree at the core of our framework. This structure is chosen for its ability to select high-level tasks based on environmental triggers, and to extend to new situations as they arise. We evaluate the system in 105 trials on the Toyota Human Support Robot, across five hardware and three simulation scenarios. These trials capture the decisions the robot must make in this setting: whether to press a door button, yield to a nearby person, walk through a door someone else is holding, or first retrieve rescue equipment before traversing the door. Overall, the system completes 97 of the 105 trials successfully. These results suggest our framework provides a practical basis for robotic assistance in broader emergency response tasks. Code and video demonstrations are available at https://github.com/AndrewSnowdy/hsr_mm_control.

Figures

Figures reproduced from arXiv: 2607.05315 by the authors.

**Figure 2.** Overview of the proposed modular architecture. The Perception module builds a LiDAR costmap and localizes the robot. It also detects task-relevant [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Real-time perception using YOLOv11n. The system identifies ADA [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: The behavior tree used by the executive. Selector and Sequence [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Sequential frames from the five hardware scenarios: nominal traversal (S1), yielding to a person crossing the doorway (S2), passing through a [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: Sequential frames from the three simulation scenarios: yielding to a simulated person crossing the doorway (S6), navigating around a static obstacle [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 4 linked inside Pith

[1]

After action report: Gabriel house fire,

Fall River Fire Department, “After action report: Gabriel house fire,” https://cms2.revize.com/revize/fallriver/FRFD%20AFTER% 20ACTION%20REPORT-%20GABRIEL%20HOUSE%20FIRE.pdf, 2025, accessed 2025-06-27

2025
[2]

Colledanchise and P

M. Colledanchise and P. ¨Ogren,Behavior Trees in Robotics and AI: An Introduction. CRC Press, 2018. [Online]. Available: https://arxiv.org/abs/1709.00084

arXiv 2018
[3]

BehaviorTree.CPP: A high-performance and generic Be- havior Tree library,

D. Faconti, “BehaviorTree.CPP: A high-performance and generic Be- havior Tree library,”GitHub repository, 2021

2021
[4]

Planning for autonomous door opening with a mobile manipulator,

S. Chitta, B. Cohen, and M. Likhachev, “Planning for autonomous door opening with a mobile manipulator,” in2010 IEEE International Conference on Robotics and Automation. IEEE, 2010, pp. 1799–1806

2010
[5]

Robust and adaptive door operation with a mobile robot,

M. Arduengo, F. Moreno-Noguer, and G. Aleny `a, “Robust and adaptive door operation with a mobile robot,”Intelligent Service Robotics, vol. 14, no. 3, pp. 409–423, 2021. [Online]. Available: https://arxiv.org/abs/1902.09051

arXiv 2021
[6]

After you: Doorway negotiation for human-robot and robot-robot interaction,

R. Messing, D. R. Thompson, and S. Chien, “After you: Doorway negotiation for human-robot and robot-robot interaction,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 1–9

2018
[7]

Chance- constrained convex mpc for robust quadruped locomotion under para- metric and additive uncertainties,

A. Trivedi, S. Prajapati, M. Zolotas, M. Everett, and T. Padır, “Chance- constrained convex mpc for robust quadruped locomotion under para- metric and additive uncertainties,”IEEE Robotics and Automation Letters, 2025

2025
[8]

Versatile multicontact planning and control for legged loco-manipulation,

J.-P. Sleiman, F. Farshidian, and M. Hutter, “Versatile multicontact planning and control for legged loco-manipulation,”Science Robotics, vol. 8, no. 81, p. eadg5014, 2023

2023

Show all 14 references

[9]

Learning to open and traverse doors with a legged manipulator,

M. Zhang, Y . Ma, T. Miki, and M. Hutter, “Learning to open and traverse doors with a legged manipulator,”arXiv preprint arXiv:2409.04882, 2024

2024 arXiv
[10]

A behavior architecture for fast humanoid robot door traversals,

D. Calvert, L. Penco, D. Anderson, T. Bialek, A. Chatterjee, B. Mishra, G. Clark, S. Bertrand, and R. Griffin, “A behavior architecture for fast humanoid robot door traversals,”Robotics and Autonomous Systems, p. 105217, 2025

2025
[11]

I-yolov11n: a lightweight and efficient small target detection framework for uav aerial images,

Y . Ma, C. Xi, T. Ma, H. Sun, H. Lu, X. Xu, and C. Xu, “I-yolov11n: a lightweight and efficient small target detection framework for uav aerial images,”Sensors, vol. 25, no. 15, p. 4857, 2025

2025
[12]

Introduction to inverse kinematics with jacobian transpose, pseudoinverse and damped least squares methods,

S. R. Buss, “Introduction to inverse kinematics with jacobian transpose, pseudoinverse and damped least squares methods,”IEEE Journal of Robotics and Automation, vol. 17, no. 1-19, p. 16, 2004

2004
[13]

Interpolation by integro quintic splines,

H. Behforooz, “Interpolation by integro quintic splines,”Applied mathematics and computation, vol. 216, no. 2, pp. 364–367, 2010

2010
[14]

A comparative study of bug algorithms for robot navigation,

H. G. de Marina, Y . A. Kapitanyuk, M. Egerstedt, G. Hu, and M. Cao, “A comparative study of bug algorithms for robot navigation,”Robotics and Autonomous Systems, 2018. [Online]. Available: https://arxiv.org/abs/1808.05050

2018 arXiv

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.

[1] [1]

After action report: Gabriel house fire,

Fall River Fire Department, “After action report: Gabriel house fire,” https://cms2.revize.com/revize/fallriver/FRFD%20AFTER% 20ACTION%20REPORT-%20GABRIEL%20HOUSE%20FIRE.pdf, 2025, accessed 2025-06-27

2025

[2] [2]

Colledanchise and P

M. Colledanchise and P. ¨Ogren,Behavior Trees in Robotics and AI: An Introduction. CRC Press, 2018. [Online]. Available: https://arxiv.org/abs/1709.00084

arXiv 2018

[3] [3]

BehaviorTree.CPP: A high-performance and generic Be- havior Tree library,

D. Faconti, “BehaviorTree.CPP: A high-performance and generic Be- havior Tree library,”GitHub repository, 2021

2021

[4] [4]

Planning for autonomous door opening with a mobile manipulator,

S. Chitta, B. Cohen, and M. Likhachev, “Planning for autonomous door opening with a mobile manipulator,” in2010 IEEE International Conference on Robotics and Automation. IEEE, 2010, pp. 1799–1806

2010

[5] [5]

Robust and adaptive door operation with a mobile robot,

M. Arduengo, F. Moreno-Noguer, and G. Aleny `a, “Robust and adaptive door operation with a mobile robot,”Intelligent Service Robotics, vol. 14, no. 3, pp. 409–423, 2021. [Online]. Available: https://arxiv.org/abs/1902.09051

arXiv 2021

[6] [6]

After you: Doorway negotiation for human-robot and robot-robot interaction,

R. Messing, D. R. Thompson, and S. Chien, “After you: Doorway negotiation for human-robot and robot-robot interaction,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 1–9

2018

[7] [7]

Chance- constrained convex mpc for robust quadruped locomotion under para- metric and additive uncertainties,

A. Trivedi, S. Prajapati, M. Zolotas, M. Everett, and T. Padır, “Chance- constrained convex mpc for robust quadruped locomotion under para- metric and additive uncertainties,”IEEE Robotics and Automation Letters, 2025

2025

[8] [8]

Versatile multicontact planning and control for legged loco-manipulation,

J.-P. Sleiman, F. Farshidian, and M. Hutter, “Versatile multicontact planning and control for legged loco-manipulation,”Science Robotics, vol. 8, no. 81, p. eadg5014, 2023

2023