Organizations: Spacetime Intelligence Laboratory, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea. · Systems Intelligence Laboratory (SILAB), KAIST, Daejeon, South Korea. · Brighton Ai Group, Brisbane, QLD, Australia.
Pedestrians increasingly encounter robots in urban public spaces without interacting with them, but how robots change the perceived safety of a street remains unclear. We examine how robot presence and type affect perceived safety across urban contexts. We generate 18 images of Seoul streets that vary in time of day, road class, and robot condition (no robot, a delivery robot, or a humanoid). Two multimodal large language models (MLLMs), Qwen3.8-27B and Gemma4-26B, each conditioned on 500 synthetic personas, choose the safer image in each pair, and 25 human respondents complete the same task. Both models prefer streets without robots and favor the delivery robot over the humanoid. Robots are chosen more often at night and less often by older personas, whereas road class has smaller effects. In the model rationales, the delivery robot appears as a physical obstacle, whereas the humanoid appears as either a threat or a source of protection. Like humans, both models chose robots more often at night, but chose the humanoid over the delivery robot far less often, and Gemma showed a night effect for the humanoid about three times that of humans. These results show that the effect of a robot on perceived safety depends on when it appears, and that persona-conditioned MLLMs capture only part of human judgments.
Figures & tables
Fig. 1 : Win rate by urban contexts. Rows show the two conditions: (a) time of day and (b) road class, and columns show robot pairs.
Fig. 2 : Win rate by persona attribute. Rows show the four attributes: (a) gender, (b) age, (c) income, and (d) region, and columns show robot pairs.
Fig. 3 : Share of rationale topics stated by Gemma4-26B for each choice, by (a) time of day and (b) persona gender, grouped by whether they describe the robot as a risk or a help, or the street. Shares within each bar sum to 100%, and numbers to the right give the number of responses.
Fig. 4 : Win rate of humans and matched personas. Rows show (a) time of day and (b) gender, and columns show robot pairs. Model results use personas matched to the human respondents by gender and age group.
Qwen3.8-27B
Gemma4-26B
Robot pair
Same
Swapped
Same
Swapped
No robot vs. Delivery
65%
52%
71%
61%
No robot vs. Humanoid
84%
77%
81%
73%
Delivery vs. Humanoid
93%
93%
94%
92%
All pairs
81%
74%
82%
75%
TABLE I: Response consistency of the two models. Each value shows the proportion of repeated trials where the model chose the same image again. The two images were presented in either the same or swapped positions.
Appendix figures & tables5 assets
Supplementary material from the paper’s appendix.
Appendix
Fig. 5 : The 18 stimulus images. Rows show the three road classes by day and at night, and columns show robot conditions. Each trial compared two images in the same row.
Attribute
Level
All
Matched
Gender
Female
250
75
Male
250
62
Age group
20s or younger
84
74
30s
84
39
40s
83
0
50s
83
24
Appendix
TABLE II: Composition of the personas. All shows the 500 personas used for the model results, and Matched shows the 137 unique personas matched to the human respondents.
Topic
Annotator
Claude
κ
(1) Collision or path obstruction
70
71
0.83
(2) Threat or unfamiliarity
100
104
0.82
(3) Openness, visibility, or lighting
20
23
0.81
(4) Surveillance or protection
33
29
0.92
(5) Robot function or trust in technology
0
5
–
All topics
223
232
0.87
Appendix
TABLE III: Agreement between the human annotator and Claude for each topic. Counts give the number of rationales that each coder assigned to the topic and sum to more than 150 because a rationale could receive more than one topic.
Fig. 6 : Share of rationale topics stated by Qwen3.8-27B for each choice, by (a) time of day and (b) persona gender. Shares within each bar sum to 100%, and numbers to the right give the number of responses.
Fig. 7 : Effect sizes for humans and matched personas. Rows show (a) time of day and (b) gender, and columns show robot pairs. Each value is Cohen’s h for the change in the win rate of the underlined target condition, from day to night or from male to female. Error bars show 95% confidence intervals, and the dashed line marks no effect.
Social robotics studies robots that interact with people through social communication, assistance, collaboration, and behavior adapted to human contexts.
This paper presents an explainable approach to pedestrian routing, in which perceived safety is estimated from street-level imagery through an explicit natural-language intermediate representation. A vision--language model caption is generated and stored before any scoring is undertaken, and the perceived-risk class is derived entirely from structured features of that stored text, so that every segment score remains inspectable by the user. Nine captioning conditions across five model families are benchmarked against a direct Contrastive Language--Image Pre-training (CLIP) image-embedding baseline under an identical downstream pipeline, and the caption-mediated representation is found to reach parity with the image embedding rather than to trail it. The approach was deployed over 654,115 images covering 36 electoral wards in two locations in Northern England (Manchester and Huddersfield). Independent field validation against 3,669 locally collected ratings of 494 images across 70 participant sessions established agreement that is statistically significant but modest, at r=0.262, against a measured noise ceiling of 0.737 imposed by disagreement between raters. A single-use confirmatory test then found that a pipeline 44% stronger on the supervised benchmark did not produce measurable improvement in the field (r=0.250, p=0.84), so the benchmark gains did not predict the deployment gains in this case. Routing behaviour varies systematically with journey length. There is negligible change below 1,km, reaching a median increase of 12.78% in low-risk route length for a median detour of 2.73% on journeys of 3 to 6 km.
Simon Parkinson, Paloma Liu, Wei Zheng +1
School of Computing and Mathematics, Manchester Metropolitan University, Manchester M1 5GD, U.K. · University of Huddersfield, Huddersfield HD1 3DH, U.K. · The University of Manchester, Manchester M13 9PL, U.K.
This study examines how persona prompting shapes language generated by two multimodal large language models in urban perception, a setting for examining subjective interpretations of shared visual evidence. We organize outputs into three functional layers: descriptive grounding (captions), intermediate semantic layer (perception tags), and interpretive framing (justifications). Using approximately 60,000 persona-conditioned annotations from each of two MLLMs, Qwen3-VL and Gemma4, we find that captions converge strongly across persona profiles and show only small attribute-associated differences. Justifications vary substantially more: economic status produces the largest difference in both models, with political orientation and personality also prominent. Paired image-level comparisons confirm larger justification than caption differences for these three attributes. For perception tags, personas sharing the same attribute level produce more similar tag sets than personas with different attribute levels, with the largest separation observed for economic status. Exploratory topic analysis further suggests persona-specific evaluative emphasis. Across models, profile-pair similarity patterns are strongly correlated for all three output types, although agreement is lowest for justifications. Overall, persona prompting affects interpretive framing more strongly than descriptive grounding.
Neemias da Silva, Matt Ratto, Myriam Delgado +3
University of Toronto, Toronto, Canada · Universidade Tecnologica Federal do Parana, Curitiba, Brazil
Large Language Models (LLMs) are increasingly used as proxies for human perception in urban analysis, yet it remains unclear whether persona prompting produces meaningful and reproducible behavioral diversity. We investigate whether distinct personas influence urban sentiment judgments generated by multimodal LLMs. Using a factorial set of personas spanning gender, economic status, political orientation, and personality, we instantiate multiple agents per persona to evaluate urban scene images from the PerceptSent dataset and assess both within-persona consistency and cross-persona variation. Results show strong convergence among agents sharing a persona, indicating stable and reproducible behavior. However, cross-persona differentiation is limited: economic status and personality induce statistically detectable but practically modest variation, while gender shows no measurable effect and political orientation only negligible impact. Agents also exhibit an extremity bias, collapsing intermediate sentiment categories common in human annotations. As a result, performance remains strong on coarse-grained polarity tasks but degrades as sentiment resolution increases, suggesting that simple label-based persona prompting does not capture fine-grained perceptual judgments. To isolate the contribution of persona conditioning, we additionally evaluate the same model without personas. Surprisingly, the no-persona model sometimes matches or exceeds persona-conditioned agreement with human labels across all task variants, suggesting that simple label-based persona prompting may add limited annotation value in this setting.
Neemias B da Silva, Rodrigo Minetto, Daniel Silver +1
Universidade Tecnol´ogica Federal do Paran´a (UTFPR), Curitiba, Brazil · University of Toronto, Toronto, Canada