Improving Human-Robot Teamwork in Urban Search and Rescue Through Episodic Memory of Prior Collaboration

Emma van Zoelen; Mark Neerincx; Taewoon Kim

REVIEW 3 major objections 3 minor 34 references

Improving Human-Robot Teamwork in Urban Search and Rescue Through Episodic Memory of Prior Collaboration

T0 review · 3 major / 3 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A robot that starts with one stored collaboration pattern rescues more victims faster.

desk verdict The headline improvement (25.7% to 41.3%) rests on a non-concurrent baseline and no inferential statistics, so the causal claim in the abstract is not supported; the pipeline itself is a concrete, honestly-reported contribution that deserves a serious referee but not publication as-is. read the letter →

arxiv 2606.18836 v2 pith:5DCRSZYN submitted 2026-06-17 cs.HC cs.AI

Taewoon Kim , Emma van Zoelen , Mark Neerincx This is my paper · ORCID

classification cs.HCcs.AI

keywords human-robotteamworkepisodicmemoryknowledgegraphcollaborationpatternurbansearchandrescueneuralnetworktransferlearninghuman-subjectevaluation

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 a robot entering a new human-robot team performs better if it starts with one concrete memory of how a previous team worked together, rather than an empty memory. In the MATRX urban search and rescue simulation, the authors preload the robot with a single collaboration pattern selected from 209 prior patterns, stored as a knowledge graph. Across 160 round-level observations from 20 participants, rescue success rises from 25.7% to 41.3% and average task time drops by 283 seconds, with the largest gains in the first round. The claim matters because it points to a cheap intervention—reusing one inspectable situation-action routine—that improves early coordination before teammates have adapted to each other.

What carries the argument

The central object is the collaboration pattern (CP): a human-readable situation plus an ordered sequence of robot and human actions, stored as a knowledge-graph episodic memory. The selection pipeline uses an RGCN with a node-classification objective to embed each CP graph so that action order is preserved (nine relation labels distinguish early from later actions), K-means clustering to group structurally similar patterns, and a centroid-nearest exemplar as the preloaded memory. That preload lets the robot reuse a known situation-action routine at the start of a new episode instead of starting from an empty memory.

What would settle it

A pre-registered study that randomly assigns participants to memory-initialized or empty-memory robots in the same sessions, with identical instructions and interface, and analyzes rescue success at the participant level: if the 41.3% versus 25.7% gap does not reproduce, the central claim is not supported.

Watch

Extended reading notes

Core claim

The authors claim that initializing a robot with a single prior collaboration pattern—stored as a knowledge-graph episodic memory—improves a new human-robot team's performance in the MATRX urban search and rescue task. They represent 209 collaboration patterns collected from earlier sessions as small knowledge graphs, embed them with a relational graph convolutional network trained to classify node types, cluster the embeddings, and select the pattern closest to the center of the high-success, structurally simple cluster. That pattern contains only two robot actions: picking up a large rock at the top of the rubble pile and dropping it on the right side of the field, giving the robot an auto

Load-bearing premise

The load-bearing premise is that the earlier, separately-run study is a fair control; if unmeasured differences in participants, instructions, interface, or timing caused the improvement, the memory preload did not.

Editorial extensions

If this is right

Robots can begin teamwork with reusable task knowledge, and the strongest gains appear in the first round, before within-session adaptation has occurred.
A simple, robot-autonomous CP—two consecutive large-rock moves—can give the human room to focus on other tasks, improving success in most rounds.
Because the reused memory remains an explicit situation-action structure, operators could inspect, revise, or disable it before deployment, keeping the transfer under human oversight.
The benefit is not uniform: victim harm rises on average and performance drops in the hardest rounds (7–8), indicating safety-relevant trade-offs when heavy-debris actions are reused in riskier scenarios.
The same selected CP is reused for all participants, so the evaluation tests transfer of one shared prior team experience rather than per-user personalization.

Reading between the lines

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

The paper does not compare its learned selection against a randomly chosen CP, so the observed gain could stem from having any sensible default action plan rather than from the graph-learning selection; a direct comparison would isolate the selection method's contribution.
If the early-round boost comes from giving the human a predictable division of labor, a hand-authored simple plan might produce similar gains, making the representation-learning pipeline optional for practical deployment.
The early-round advantage suggests episodic memory may substitute for within-session adaptation; a testable extension is whether the benefit persists or fades in longer sessions as teams form their own routines.
The increased victim harm in later rounds implies that memory reuse should be gated by situational risk—for example, preloading multiple CPs with situation-triggered selection rather than one fixed routine.

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

The paper's central claim—that preloading a robot with one selected collaboration pattern raises rescue success from 25.7% to 41.3%—is not supported by the evidence as reported. The baseline comes from a separately run earlier study, not a concurrent control, and the memory-initialized condition is compared to it without any inferential statistics. The abstract's causal phrasing overstates what the experiment can show.

What's actually new and good: the authors do a concrete job of representing collaboration patterns as knowledge graphs, learning graph embeddings with an RGCN, clustering them, selecting a centroid-nearest CP, and preloading it in the MATRX USAR simulator. That specific pipeline—turning reflective collaboration traces into a reusable, inspectable prior memory—is a new application relative to earlier MATRX work. They ship code and data, and they are transparent about limitations: they explicitly say the selection is heuristic rather than tuned, they note the round-level difference is not a formal participant-level comparison, and they show victim harm increases and later rounds get worse. That honesty is real and worth crediting.

The soft spots are proportional. The main one is load-bearing: no concurrent randomized control, and no participant-level modeling or confidence intervals. Section IV.C says the memory condition ran 'under the same conditions as the previous study'—that is an assertion, not demonstrated equivalence. Time, participant pool, interface versions, instructions, environment changes—any of these could explain the difference. Also, the selected CP was chosen after inspecting cluster-level success rates (Cluster 1 had the highest historical success at 42%), so the evaluation partly tests that hand-picked pattern rather than the selection method itself. The method's generality is untested: one simulated scenario, 20 participants, one preloaded CP. The free parameters (K=5, cluster selection) are acknowledged as heuristics, which is fine, but they don't strengthen the causal evidence.

This isn't a takedown. The pipeline is plausible and the authors are honest. But the empirical foundation for the headline claim is weak. The fix is straightforward: run a concurrent randomized experiment with the same participant pool, analyze at the participant level (e.g., mixed-effects models), and ideally compare the selected CP against a random CP or no-memory condition. Until then, the effect size is suggestive, not demonstrated.

Who is this paper for? Researchers in HRI memory transfer or MATRX co-learning. It demonstrates one clean way to operationalize 'prior team experience' as an explicit graph memory. A serious referee could help the authors redesign the evaluation. I'd recommend sending it to peer review—it is not desk-rejectable—but with a clear expectation that the empirical claims will need major revisions or substantial re-scoping.

Referee Report

3 major / 3 minor

Summary. The paper proposes representing collaboration patterns (CPs) from earlier MATRX USAR studies as knowledge-graph episodic memories, using an RGCN with a node-classification objective to embed and cluster these CPs, and then selecting one centroid-nearest CP from a cluster judged to be effective. The selected CP is preloaded into the robot's memory before a new human-robot session. The authors report, across 20 participants and 160 round-level observations, an increase in rescue success from 25.7% to 41.3% and a reduction in average task time by 283 seconds, with the largest gains in early rounds. The contribution is positioned as a mechanism for pre-interaction transfer of inspectable prior team experience.

Significance. If the reported effect were well-supported, the work would be a valuable step toward using explicit episodic memory to improve early human-robot teamwork; the memory representation is human-readable, and the authors provide code and data. The evaluation on new participants is a positive feature. However, the central empirical claim is currently undermined by the lack of a concurrent control condition and by the absence of participant-level inferential statistics, so the significance of the result cannot be assessed.

major comments (3)

[Section IV.C / Table V] The central comparison (25.7% to 41.3% success, 283s time reduction) is between the memory-initialized condition and a baseline from a previously run study. The paper asserts the experiments were run under the same conditions with the only difference being the preloaded memory, but no evidence is provided about participant-pool equivalence, interface versions, instructions, or time-related drift. Because the baseline is non-concurrent and non-randomized, unmeasured confounds could fully account for the observed difference. This is the load-bearing comparison for the abstract's causal claim; please either add a concurrent randomized baseline or reframe the claims as observational without causal language.
[Section IV.D] The paper correctly notes that the 160 round-level observations are not independent, but then does not supply the needed participant-level analysis. No confidence intervals, effect sizes, or significance tests are reported for the success-rate or time comparisons. As a result, the magnitude and even the existence of the effect is unquantified. A participant-level summary (e.g., mean per-participant success, paired analysis across rounds) is required to support the headline numbers.
[Section IV.B / Table II] The choice of Cluster 1 was made after inspecting cluster-level success rates (Table II), where Cluster 1 had the highest historical success (42%), and the selected CP is the centroid-nearest CP of that cluster. The abstract's phrase 'automatically selected prior CP' overstates this procedure: Section IV.B describes heuristic selection with manual inspection. This outcome-informed selection can bias the apparent transfer benefit, and there is no comparison to a randomly selected CP or an expert-designed simple pattern. Please present the selection as exploratory or pre-register the selection rule, and add a sensitivity analysis using a non-selected CP.

minor comments (3)

[Section III.B] The node-classification objective is a pretext task; the paper does not validate that the embeddings improve clustering quality over raw features beyond visual inspection in Figure 6. A quantitative cluster-quality comparison would strengthen the method claims.
[Figure 5] Training loss and accuracy are reported without validation. State whether any regularization or early stopping was used, and consider reporting validation performance.
[Table I / Table IV] Object labels vary between 'Large rock' (Table IV) and 'Brown rock' (Table I and Section IV.D). Clarify whether these refer to the same object type and how the selected CP interacts with brown-rock rounds.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found; the central result is an out-of-sample experimental comparison, not a fitted parameter recycled as a prediction.

full rationale

The derivation chain is not circular. The RGCN is trained with a node-type classification objective, not with the target success metric, and the paper explicitly states that the learned embeddings are used only to group structurally similar CPs during heuristic selection, 'not to directly rank memories by their expected transfer value.' The selected CP is the centroid-nearest member of Cluster 1; although Cluster 1 was chosen partly because of its historical success rate, the evaluation of the memory-initialized robot is conducted on 20 new participants, providing an out-of-sample test. The reported improvement (25.7% to 41.3%) is an observed experimental difference, not a fitted parameter renamed as a prediction. The main validity threat is that the control condition is taken from an earlier study under an asserted 'same conditions' equivalence rather than a concurrent randomized control, and the paper itself acknowledges 'lack of random, expert-selected, or multi-CP preload baselines.' However, non-concurrent baselines and small-sample limitations are experimental-design concerns, not circularity by construction. No self-citation chain is load-bearing for the central claim; prior MATRX work is cited for the environment and CP ontology only, not to derive the result. Therefore no quoted step reduces to its own inputs.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a handful of domain and design assumptions about the MATRX simulation, the meaning of participant-defined CPs, and the comparability of historical baseline data. No new entities are postulated. Free parameters are limited to heuristic choices (K=5, manual cluster selection) that influence the result.

free parameters (2)

Number of clusters K = 5
Chosen by exploratory inspection of cluster interpretability and balance; not optimized against downstream performance. Directly affects which CP is selected and thus the reported improvement.
Cluster 1 selection heuristic = Cluster 1 (42% success rate, simple structure)
Manual choice based on cluster success rate and structural simplicity (Section IV.B). Not part of a fixed, pre-specified selection rule; uses outcome information from the historical data.

assumptions (5)

domain assumption The MATRX USAR simulation is a valid proxy for real urban search and rescue teamwork.
The entire study is conducted in a virtual environment; generalization to physical disaster response is assumed without validation (Section II.A).
domain assumption Participant-defined collaboration patterns are meaningful, reusable routines that capture effective team behavior.
CPs are self-reported, variable, and may be idiosyncratic; the paper treats them as reliable episodic memories without inter-annotator or validity checks (Sections II.A, III.A).
ad hoc to paper The RGCN node-classification objective produces embeddings that separate useful from less useful CPs for transfer.
The node classes are schema-defined (action order, entity types), not outcome-defined. The paper does not validate that embedding proximity corresponds to transfer value; it is a heuristic convenience (Section III.B).
ad hoc to paper The historical baseline is comparable to the new condition except for the memory preload.
Section IV.C assumes 'the same conditions as the previous study.' This comparability is not tested and is central to the causal interpretation of the improvement.
domain assumption A single preloaded CP will be reused appropriately by the existing robot control framework when the situation matches, without adverse interactions.
The execution mechanism is described only as 'the robot can reuse the preloaded CP when the current situation matches the stored condition' (Section III.B). The matching and execution details are unspecified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Improving Human-Robot Teamwork in Urban Search and Rescue Through Episodic Memory of Prior Collaboration." pith.science (2026). https://pith.science/paper/5DCRSZYN

@misc{pith2026260618836,
  author       = {Pith},
  title        = {Pith review of: Improving Human-Robot Teamwork in Urban Search and Rescue Through Episodic Memory of Prior Collaboration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5DCRSZYN}},
  note         = {Machine review of arXiv:2606.18836}
}

read the original abstract

Effective human-robot teamwork requires robots to adapt to partners, situations, and task dynamics from the start of an interaction. In the MATRX Urban Search and Rescue (USAR) environment, people can externalize collaboration patterns (CPs) they discover during teamwork through a chat and reflection interface. We study whether a robot can use such prior team experience to become a better teammate in future interactions. To this end, we represent historical CPs as knowledge-graph episodic memories and use graph representation learning with a node-classification objective to identify a representative and effective memory for reuse. We then initialize the robot with this memory before a new collaboration episode begins. Across 20 participants and 160 round-level observations, initializing the robot with a single automatically selected prior CP increases rescue success from 25.7% to 41.3% and reduces average task time by 283 seconds. The strongest gains appear at the beginning of interaction, suggesting that reusable episodic memory can help robots enter collaboration with more effective task knowledge and support smoother early teamwork.

Figures

Figures reproduced from arXiv: 2606.18836 by the authors.

**Figure 2.** User interaction features in the MATRX USAR [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Visualizations of collaboration patterns (CPs) rep [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: A forward pass of the RGCN-based neural network. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 5.** Figure 5: Cross entropy loss and accuracy of the training data. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 6.** Figure 6: K-means clustering results before and after training, [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 2 linked inside Pith

[1]

Record of 4 million robots working in factories worldwide,

IFR, “Record of 4 million robots working in factories worldwide,”
[2]

Collaborative robot market by payload, compo- nent, industry, and region - global forecast to 2030,

MarketsandMarkets, “Collaborative robot market by payload, compo- nent, industry, and region - global forecast to 2030,” 2024. Accessed: 2025-02-03
[3]

Human-robot mutual adapta- tion in collaborative tasks: Models and experiments,

S. Nikolaidis, D. Hsu, and S. Srinivasa, “Human-robot mutual adapta- tion in collaborative tasks: Models and experiments,”The International Journal of Robotics Research, vol. 36, p. 027836491769059, 02 2017

2017
[4]

Research needs in human-autonomy teaming: Thematic analysis of priority features for testbed development,

M. O. Smith, S. Amatya, A. Amresh, J. C. Gorman, M. Johnson, N. J. Cooke, and W. Zhang, “Research needs in human-autonomy teaming: Thematic analysis of priority features for testbed development,” in 33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1183–1190, IEEE, 2024

2024
[5]

Human- robot action teams: A behavioral analysis of team dynamics,

A. Haripriyan, R. Jamshad, P. Ramaraj, and L. D. Riek, “Human- robot action teams: A behavioral analysis of team dynamics,” in 33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1443–1448, IEEE, 2024

2024
[6]

Collaborative robots (cobots) for disaster risk resilience: a framework for swarm of snake robots in delivering first aid in emergency situations,

S. K. R. Moosavi, M. H. Zafar, and F. Sanfilippo, “Collaborative robots (cobots) for disaster risk resilience: a framework for swarm of snake robots in delivering first aid in emergency situations,”Frontiers in Robotics and AI, vol. 11, p. 1362294, 2024

2024
[7]

Perceptions and opinions of rescuers about a quadruped robot in an earthquake scenario,

Z. Betta, A. Gaudino, A. Benini, C. T. Recchiuto, and A. Sgorbissa, “Perceptions and opinions of rescuers about a quadruped robot in an earthquake scenario,” in33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1092–1099, IEEE, 2024

2024
[8]

On the effect of augmented-reality multi-user interfaces and shared mental models on human-robot trust,

R. Chacón-Quesada, F. E. Casado, and Y . Demiris, “On the effect of augmented-reality multi-user interfaces and shared mental models on human-robot trust,” in33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1316–1322, IEEE, 2024

2024

Show all 34 references

[9]

Matrx: Human agent teaming rapid experimentation software,

T. H. Jasper van der Waa, “Matrx: Human agent teaming rapid experimentation software,” July 2023

2023
[10]

Becoming team members: Identifying interaction patterns of mutual adaptation for human-robot co-learning,

E. van Zoelen, K. Bosch, and M. Neerincx, “Becoming team members: Identifying interaction patterns of mutual adaptation for human-robot co-learning,”Frontiers in Robotics and AI, vol. 8, 07 2021

2021
[11]

Human- robot co-learning for fluent collaborations,

E. M. van Zoelen, K. van den Bosch, and M. Neerincx, “Human- robot co-learning for fluent collaborations,” inCompanion of the 2021 ACM/IEEE International Conference on Human-Robot Interaction, HRI ’21 Companion, (New York, NY , USA), p. 574–576, Association for Computing Machi...

2021
[12]

van Zoelen, K

E. van Zoelen, K. Bosch, D. Abbink, and M. Neerincx,Ontology- Based Reflective Communication for Shared Human-AI Recognition of Emergent Collaboration Patterns, pp. 621–629. 11 2022

2022
[13]

Shared knowledge in human-robot interaction (hri),

L. Miraglia, C. D. Dio, F. Manzi, T. Kanda, A. Cangelosi, S. Itakura, H. Ishiguro, D. Massaro, P. Fonagy, and A. Marchetti, “Shared knowledge in human-robot interaction (hri),”International Journal of Social Robotics, vol. 16, p. 59–75, 2024

2024
[14]

A framework for trust-related knowledge transfer in human-robot interaction,

M. Diab and Y . Demiris, “A framework for trust-related knowledge transfer in human-robot interaction,”Autonomous Agents and Multi- Agent Systems, vol. 38, no. 1, p. 24, 2024

2024
[15]

The collaborative mind: Intention reading and trust in human-robot interaction,

S. Vinanzi, A. Cangelosi, and C. Goerick, “The collaborative mind: Intention reading and trust in human-robot interaction,”iScience, vol. 24, no. 2, p. 102130, 2021

2021
[16]

Typeql: A type-theoretic & polymorphic query language,

C. Dorn and H. Pribadi, “Typeql: A type-theoretic & polymorphic query language,”Proceedings of the ACM on Management of Data, vol. 2, pp. 1–27, 05 2024

2024
[17]

E. M. van Zoelen,Human-Machine Co-Learning: Anticipating, Identi- fying and Sharing Emergent Collaboration Patterns. PhD thesis, Delft University of Technology, 2025

2025
[18]

Knowledge graphs,

A. Hogan, E. Blomqvist, M. Cochez, and et al., “Knowledge graphs,” ACM Comput. Surv., vol. 54, jul 2021

2021
[19]

RDF 1.1 concepts and abstract syntax,

M. Lanthaler, D. Wood, and R. Cyganiak, “RDF 1.1 concepts and abstract syntax,” W3C recommendation, W3C, February 2014. https://www.w3.org/TR/2014/REC-rdf11-concepts-20140225/

2014
[20]

Janusgraph: an open-source, distributed graph database,

JanusGraph Contributors, “Janusgraph: an open-source, distributed graph database,” 2024

2024
[21]

The graph neural network model,

F. Scarselli, M. Gori, A. C. Tsoi, and et al., “The graph neural network model,”IEEE Transactions on Neural Networks, vol. 20, no. 1, pp. 61– 80, 2009

2009
[22]

Semi-supervised classification with graph convolutional networks,

T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,”CoRR, vol. abs/1609.02907, 2016

2016 arXiv
[23]

Modeling relational data with graph convolutional networks,

M. Schlichtkrull, T. N. Kipf, P. Bloem, R. van den Berg, I. Titov, and M. Welling, “Modeling relational data with graph convolutional networks,” 2017

2017
[24]

Sentence-bert: Sentence embeddings using siamese bert-networks,

N. Reimers and I. Gurevych, “Sentence-bert: Sentence embeddings using siamese bert-networks,” 2019

2019
[25]

Visualizing data using t-sne,

L. van der Maaten and G. Hinton, “Visualizing data using t-sne,” Journal of Machine Learning Research, vol. 9, pp. 2579–2605, 2008

2008
[26]

Neural episodic control,

A. Pritzel, B. Uria, S. Srinivasan, A. P. Badia, O. Vinyals, D. Hassabis, D. Wierstra, and C. Blundell, “Neural episodic control,” inPro- ceedings of the 34th International Conference on Machine Learning (D. Precup and Y . W. Teh, eds.), vol. 70 ofProceedings of Machine Learni...

2017
[27]

Efficient lifelong learning with a-gem,

A. Chaudhry, M. Ranzato, M. Rohrbach, and M. Elhoseiny, “Efficient lifelong learning with a-gem,” 2019

2019
[28]

A machine with human-like memory systems,

T. Kim, M. Cochez, V . Francois-Lavet, and et al., “A machine with human-like memory systems,” 2022

2022
[29]

A machine with short-term, episodic, and semantic memory systems,

T. Kim, M. Cochez, V . Francois-Lavet, and et al., “A machine with short-term, episodic, and semantic memory systems,”Proceedings of the Conference on Artificial Intelligence, vol. 37, Jun. 2023

2023
[30]

Temporal knowledge- graph memory in a partially observable environment,

T. Kim, V . François-Lavet, and M. Cochez, “Temporal knowledge- graph memory in a partially observable environment,” 2026

2026
[31]

Sequential transfer learning-based hu- man decision making model for human-robot co-learning and insights from user feedback analysis,

R. Kumar, V . Som, and N. Yao, “Sequential transfer learning-based hu- man decision making model for human-robot co-learning and insights from user feedback analysis,” in33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA,...

2024
[32]

Real- world human-robot collaborative reinforcement learning,

A. Shafti, J. Tjomsland, W. Dudley, and A. A. Faisal, “Real- world human-robot collaborative reinforcement learning,”CoRR, vol. abs/2003.01156, 2020

2003 arXiv
[33]

Interactive artificial intelligence: De- signing for the

J. Wenskovitch and C. North, “Interactive artificial intelligence: De- signing for the "two black boxes" problem,”Computer, vol. 53, no. 8, pp. 29–39, 2020

2020
[2023]

Accessed: 2025-02-03

2025

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Record of 4 million robots working in factories worldwide,

IFR, “Record of 4 million robots working in factories worldwide,”

[2] [2]

Collaborative robot market by payload, compo- nent, industry, and region - global forecast to 2030,

MarketsandMarkets, “Collaborative robot market by payload, compo- nent, industry, and region - global forecast to 2030,” 2024. Accessed: 2025-02-03

[3] [3]

Human-robot mutual adapta- tion in collaborative tasks: Models and experiments,

S. Nikolaidis, D. Hsu, and S. Srinivasa, “Human-robot mutual adapta- tion in collaborative tasks: Models and experiments,”The International Journal of Robotics Research, vol. 36, p. 027836491769059, 02 2017

2017

[4] [4]

Research needs in human-autonomy teaming: Thematic analysis of priority features for testbed development,

M. O. Smith, S. Amatya, A. Amresh, J. C. Gorman, M. Johnson, N. J. Cooke, and W. Zhang, “Research needs in human-autonomy teaming: Thematic analysis of priority features for testbed development,” in 33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1183–1190, IEEE, 2024

2024

[5] [5]

Human- robot action teams: A behavioral analysis of team dynamics,

A. Haripriyan, R. Jamshad, P. Ramaraj, and L. D. Riek, “Human- robot action teams: A behavioral analysis of team dynamics,” in 33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1443–1448, IEEE, 2024

2024

[6] [6]

Collaborative robots (cobots) for disaster risk resilience: a framework for swarm of snake robots in delivering first aid in emergency situations,

S. K. R. Moosavi, M. H. Zafar, and F. Sanfilippo, “Collaborative robots (cobots) for disaster risk resilience: a framework for swarm of snake robots in delivering first aid in emergency situations,”Frontiers in Robotics and AI, vol. 11, p. 1362294, 2024

2024

[7] [7]

Perceptions and opinions of rescuers about a quadruped robot in an earthquake scenario,

Z. Betta, A. Gaudino, A. Benini, C. T. Recchiuto, and A. Sgorbissa, “Perceptions and opinions of rescuers about a quadruped robot in an earthquake scenario,” in33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1092–1099, IEEE, 2024

2024

[8] [8]

On the effect of augmented-reality multi-user interfaces and shared mental models on human-robot trust,

R. Chacón-Quesada, F. E. Casado, and Y . Demiris, “On the effect of augmented-reality multi-user interfaces and shared mental models on human-robot trust,” in33rd IEEE International Conference on Robot and Human Interactive Communication, RO-MAN 2024, Pasadena, CA, USA, August 26–30, 2024, pp. 1316–1322, IEEE, 2024

2024