Previous meta-analyses show that social robots aid learning but were often limited in scope or grouped diverse control conditions together. This meta-analysis examined learning outcomes, focusing on control condition type. We retrieved 146 studies (Google Scholar and reference searches) where a physical social robot was used for training cognitive skills, comprising 183 post-test effect sizes between the robot and the control group, and 372 pre-post effect sizes. Analysis of the 78 studies with control groups indicated that robots generally improved learning, most notably when compared to a no-training control group (d = 0.75). Comparing robots to human teachers yielded an overall positive effect (d = 0.31), although effect sizes varied widely. This variability was explained by the robot’s role: robots in a co-teaching capacity showed a strong positive effect (d = 0.88), while robots replacing the teacher showed no benefit (d = −0.06). LLM-based sentiment analysis indicated that papers from outside Europe received higher positivity scores when describing the robots. We conclude that the effect size is influenced by the robot implementation and the control condition chosen.

The advent of self-driving cars has sparked discussions about eye contact in traffic, particularly due to challenges that automated vehicles face in non-verbal communication with human road users. In his 1992 book, Turn Signals Are The Facial Expressions Of Automobiles, Don Norman describes how drivers in Mexico City deliberately avoid eye contact when entering a roundabout to create uncertainty in the minds of other drivers, leading the latter to yield right of way. Norman argued that such manipulative or aggressive behavior would not be tolerated in the United States. In the present study, we tested these claims through an online survey involving 3,857 respondents from 20 countries. The results confirmed that Mexican drivers reported a higher frequency of non-speeding ‘aggressive’ violations compared to those from most other countries. Regarding eye contact in the roundabout scenario presented in the survey, national differences were found not so much in the frequency of eye contact but in the reasons behind its use. Mexican drivers tended to avoid eye contact to reduce tension or avoid conflict with other drivers. However, they also frequently reported making eye contact to assert or subtly enforce their right of way. In higher-income countries like the United States, driver-driver eye contact is often deemed unnecessary. In conclusion, our findings partially correspond with Norman's anecdote based on his experiences in 1950s Mexico City. These results may have implications for understanding the stability of traffic cultures and the challenges related to eye contact and non-verbal communication faced by developers of automated vehicles.

As automated vehicles (AVs) become increasingly popular, the question arises as to how cyclists will interact with such vehicles. This study investigated (1) whether cyclists spontaneously notice if a vehicle is driverless, (2) how well they perform a driver-detection task when explicitly instructed, and (3) how they carry out these tasks. Using a Wizard-of-Oz method, 37 participants cycled a designated route and encountered an AV multiple times in two experimental sessions. In Session 1, participants cycled the route uninstructed, while in Session 2, they were instructed to verbally report whether they detected the presence or absence of a driver. Additionally, we recorded participants’ gaze behaviour with eye-tracking and their responses in post-session interviews. The interviews revealed that 30% of the cyclists spontaneously mentioned the absence of a driver (Session 1), and when instructed (Session 2), they detected the absence and presence of the driver with 93% accuracy. The eye-tracking data showed that cyclists looked more frequently and for longer at the vehicle in Session 2 compared to Session 1. Additionally, participants exhibited intermittent sampling of the vehicle, and they looked at the area in front of the vehicle when it was far away and towards the windshield region when it was closer. The post-session interviews also indicated that participants were curious, but felt safe, and reported a need to receive information about the AV's driving state. In conclusion, cyclists can detect the absence of a driver in the AV, and this detection may influence their perception of safety. Further research is needed to explore these findings in real-world traffic conditions.

Attention bias towards social threat has been linked to loneliness and anxiety, though findings are mixed and concerns about measurement reliability persist. This study examined whether state and trait loneliness, along with personality, self-esteem, social anxiety, and life satisfaction, are associated with attention bias towards social threat images (indicating rejection or exclusion) in young adults (N = 241). AI-generated images were used to enhance control over stimulus content and category distinctions. Participants completed an eye-tracking free-viewing task comprising 40 image matrices (four images per matrix, displayed for 6000 ms). We then computed attention bias (dwell time percentage, total fixation duration percentage, and fixation count percentage) and initial orientation of attention (first fixation percentage). The attention bias measures showed adequate-to-good internal consistency (α = 0.61–0.86). No significant associations emerged between loneliness and attention to socially threatening stimuli, suggesting that heightened vigilance to social threat may not be a feature of loneliness in non-clinical young adults. However, it was found that females exhibited greater attention to social positive images, and baseline pupil diameter was associated with social anxiety. Future research should assess whether loneliness-specific attention bias is a replicable phenomenon, ideally by using an extreme-sampling approach with very lonely individuals.

This study investigated human performance in identifying AI-generated images. In a speeded forced-choice task, 255 participants viewed paired images (one real, one AI-generated by Midjourney) of standard or futuristic cars and buildings and had to identify the AI-generated one, while eye movements were recorded using an eye-tracker. Results revealed a powerful “futurism-as-artificiality” heuristic. Specifically, participants performed poorly (55% correct) when an AI-generated standard image was paired with a real futuristic image. Conversely, accuracy was high (91% correct) when the AI-generated futuristic image was paired with a real standard image. Participants’ gaze landed first on the AI-generated image more often when it depicted a futuristic design than when it depicted a standard one. The demonstrated heuristic presents a double-edged sword for information veracity: it may lead to the uncritical acceptance of AI-generated misinformation that appears conventional, while simultaneously causing real forward-thinking designs to be dismissed as fake.

Recent advancements in AI have accelerated the evolution of versatile robot designs. Chess provides a standardized environment for evaluating the impact of robot behavior on human behavior. This article presents an open-source chess robot for human-robot interaction research, specifically focusing on verbal and non-verbal interactions. The OpenChessRobot recognizes chess pieces using computer vision, executes moves, and interacts with the human player through voice and robotic gestures. We detail the software design, provide quantitative evaluations of the efficacy of the robot, and offer a guide for its reproducibility. An online survey examining people’s views of the robot in three possible scenarios was conducted with 597 participants. The robot received the highest ratings in the robotics education and the chess coach scenarios, while the home entertainment scenario received the lowest scores. The code is accessible on GitHub: https://github.com/renchizhhhh/OpenChessRobot.

Robots are becoming more capable and can autonomously perform tasks such as navigating between locations. However, human oversight remains crucial. This study compared two touchless methods for directing mobile robots: voice control and gesture control, to investigate the efficiency of these methods and the preference of users. We tested these methods in two conditions: one in which participants remained stationary and one in which they walked freely alongside the robot. We hypothesized that walking alongside the robot would result in higher intuitiveness ratings and improved task performance, based on the idea that walking promotes spatial alignment and reduces the effort required for mental rotation. In a 2×2 within-subject design, 218 participants guided the quadruped robot Spot along a circuitous route with multiple 90° turns using rotate left, rotate right, and walk forward commands. After each trial, participants rated the intuitiveness of the command mapping, while post-experiment interviews were used to gather the participants’ preferences. Results showed that voice control combined with walking with Spot was the most favored and intuitive, whereas gesture control while standing caused confusion for left/right commands. Nevertheless, 29% of participants preferred gesture control, citing increased task engagement and visual congruence as reasons. An odometry-based analysis revealed that participants often followed behind Spot, particularly in the gesture control condition, when they were allowed to walk. In conclusion, voice control with walking produced the best outcomes. Improving physical ergonomics and adjusting gesture types could make gesture control more effective.

This paper proposes a novel approach to measuring human driving performance by using the AI capabilities of automated driving systems, illustrated through three example scenarios. Traditionally, the assessment of human driving has followed a bottom-up methodology, where raw data are compared to fixed thresholds, yielding indicators such as the number of hard braking events. However, acceleration threshold exceedances are often heavily influenced by the driving context. We propose a top-down context-aware approach to driving assessments, in which recordings of human-driven vehicles are analyzed by an automated driving system. By comparing the human driver’s speed to the AI’s recommended speed, we derive a level of disagreement that can be used to distinguish between hard braking caused by aggressive driving and emergency braking in response to a critical event. The proposed method may serve as an alternative to the metrics currently used by some insurance companies and may serve as a template for future AI-based driver assessment.

The processes underlying human cognition are often divided into System 1, which involves fast, intuitive thinking, and System 2, which involves slow, deliberate reasoning. Previously, large language models were criticized for lacking the deeper, more analytical capabilities of System 2. In September 2024, OpenAI introduced the o1 model series, designed to handle System 2-like reasoning. While OpenAI’s benchmarks are promising, independent validation is still needed. In this study, we tested the o1-preview model twice on the Dutch ‘Mathematics B’ final exam. It scored a near-perfect 76 and 74 out of 76 points. For context, only 24 out of 16,414 students in the Netherlands achieved a perfect score. By comparison, the GPT-4o model scored 66 and 62 out of 76, well above the Dutch students’ average of 40.63 points. Neither model had access to the exam figures. Since there was a risk of model contamination (i.e., the knowledge cutoff for o1-preview and GPT-4o was after the exam was published online), we repeated the procedure with a new Mathematics B exam that was published after the cutoff date. The results again indicated that o1-preview performed strongly (97.8th percentile), which suggests that contamination was not a factor. We also show that there is some variability in the output of o1-preview, which means that sometimes there is ‘luck’ (the answer is correct) or ‘bad luck’ (the output has diverged into something that is incorrect). We demonstrate that the self-consistency approach, where repeated prompts are given and the most common answer is selected, is a useful strategy for identifying the correct answer. It is concluded that while OpenAI’s new model series holds great potential, certain risks must be considered.

Many collisions between pedestrians and cars are caused by poor visibility, such as occlusion by a parked vehicle. Augmented reality (AR) could help to prevent this problem, but it is unknown to what extent the augmented information needs to be embedded into the world. In this virtual reality experiment with a head-mounted display (HMD), 28 participants were exposed to AR designs, in a scenario where a vehicle approached from behind a parked vehicle. The experimental conditions included a head-locked live video feed of the occluded region, meaning it was fixed in a specific location within the view of the HMD (VideoHead), a world-locked video feed displayed across the street (VideoStreet), and two conformal diminished reality designs: a see-through display on the occluding vehicle (VideoSeeThrough) and a solution where the occluding vehicle has been made semi-transparent (TransparentVehicle). A Baseline condition without augmented information served as a reference. Additionally, the VideoHead and VideoStreet conditions were each tested with and without the addition of a guiding arrow indicating the location of the approaching vehicle. Participants performed 42 trials, 6 per condition, during which they had to hold a key when they felt safe to cross. The keypress percentages and responses from additional questionnaires showed that the diminished-reality TransparentVehicle and VideoSeeThrough designs came out most favourably, while the VideoHead solution caused some discomfort and dissatisfaction. An analysis of head yaw angle showed that VideoHead and VideoStreet caused divided attention between the screen and the approaching vehicle. The use of guiding arrows did not contribute demonstrable added value. AR designs with a high level of local embeddedness are beneficial for addressing occlusion problems when crossing. However, the head-locked solutions should not be immediately dismissed because, according to the literature, such solutions can serve tasks where a salient warning or instruction is beneficial.

Marie Skłodowska-Curie Actions; Innovative Training Networks (ITN); SHAPE-IT; Grant number 860410

Publication date: 27 May 2024

DOI: 10.1007/s10055-024-01017-9

Introduction: Communication from automated vehicles (AVs) to pedestrians using augmented reality (AR) could positively contribute to traffic safety. However, previous AR research for pedestrians was mainly conducted through online questionnaires or experiments in virtual environments instead of real ones. Methods: In this study, 28 participants conducted trials outdoors with an approaching AV and were supported by four different AR interfaces. The AR experience was created by having participants wear a Varjo XR-3 headset with see-through functionality, with the AV and AR elements virtually overlaid onto the real environment. The AR interfaces were vehicle-locked (Planes on vehicle), world-locked (Fixed pedestrian lights, Virtual fence), or head-locked (Pedestrian lights HUD). Participants had to hold down a button when they felt it was safe to cross, and their opinions were obtained through rating scales, interviews, and a questionnaire. Results: The results showed that participants had a subjective preference for AR interfaces over no AR interface. Furthermore, the Pedestrian lights HUD was more effective than no AR interface in a statistically significant manner, as it led to participants more frequently keeping the button pressed. The Fixed pedestrian lights scored lower than the other interfaces, presumably due to low saliency and the fact that participants had to visually identify both this AR interface and the AV. Discussion: In conclusion, while users favour AR in AV-pedestrian interactions over no AR, its effectiveness depends on design factors like location, visibility, and visual attention demands. In conclusion, this work provides important insights into the use of AR outdoors. The findings illustrate that, in these circumstances, a clear and easily interpretable AR interface is of key importance.

Marie Skłodowska-Curie Actions; Innovative Training Networks (ITN); SHAPE-IT; Grant number 860410

Publication date: 30 January 2024

DOI: 10.3389/frobt.2024.1324060

There is a need to improve the validity of the driving test as a measure of an individual’s ability to drive safely. This paper explores the use of algorithms to analyze acceleration and GPS data from a smartphone and a GoPro to distinguish between different driving styles, as performed by experienced examiners portraying stereotypical driving test candidates. Measures from nine driving tests were analyzed, including (harsh) accelerations, jerk, mean speed, and speeding. Results showed that the type of car, instructed driving style, and driving route impacted the dependent measures. It is concluded that GPS and accelerometer data can effectively distinguish between cautious, normal, and aggressive driving. However, it is important to consider additional sensors, such as cameras, to allow for more context-aware assessments of driving behavior. Furthermore, we demonstrate methods to quantify variations in road conditions and provide suggestions for presenting the data to driving examiners.

Vision-language models are of interest in various domains, including automated driving, where computer vision techniques can accurately detect road users, but where the vehicle sometimes fails to understand context. This study examined the effectiveness of GPT-4V in predicting the level of 'risk' in traffic images as assessed by humans. We used 210 static images taken from a moving vehicle, each previously rated by approximately 650 people. Based on psychometric construct theory and using insights from the self-consistency prompting method, we formulated three hypotheses: (i) repeating the prompt under effectively identical conditions increases validity, (ii) varying the prompt text and extracting a total score increases validity compared to using a single prompt, and (iii) in a multiple regression analysis, the incorporation of object detection features, alongside the GPT-4V-based risk rating, significantly contributes to improving the model's validity. Validity was quantified by the correlation coefficient with human risk scores, across the 210 images. The results confirmed the three hypotheses. The eventual validity coefficient was r = 0.83, indicating that population-level human risk can be predicted using AI with a high degree of accuracy. The findings suggest that GPT-4V must be prompted in a way equivalent to how humans fill out a multi-item questionnaire.

Introduction: Despite their important role in the economy, truck drivers face several challenges, including adapting to advancing technology. The current study investigated the occupational experiences of Dutch truck drivers to detect common patterns. Methods: A questionnaire was distributed to professional drivers in order to collect data on public image, traffic safety, work pressure, transport crime, driver shortage, and sector improvements. Results: The findings based on 3,708 respondents revealed a general dissatisfaction with the image of the industry and reluctance to recommend the profession. A factor analysis of the questionnaire items identified two primary factors: ‘Work Pressure’, more common among national drivers, and ‘Safety & Security Concerns’, more common among international drivers. A ChatGPT-assisted analysis of textbox comments indicated that vehicle technology received mixed feedback, with praise for safety and fuel-efficiency improvements, but concerns about reliability and intrusiveness. Discussion: In conclusion, Dutch professional truck drivers indicate a need for industry improvements. While the work pressure for truck drivers in general may not be high relative to certain other occupational groups, truck drivers appear to face a deficit of support and respect.

Personality research has traditionally relied on questionnaires, which bring with them inherent limitations, such as response style bias. With the emergence of large language models such as ChatGPT, the question arises as to what extent these models can be used in personality research. In this study, ChatGPT (GPT-4) generated 2000 text-based personas. Next, for each persona, ChatGPT completed a short form of the Big Five Inventory (BFI-10), the Brief Sensation Seeking Scale (BSSS), and a Short Dark Triad (SD3). The mean scores on the BFI-10 items were found to correlate strongly with means from previously published research, and principal component analysis revealed a clear five-component structure. Certain relationships between traits, such as a negative correlation between the age of the persona and the BSSS score, were clearly interpretable, while some other correlations diverged from the literature. An additional analysis using four new sets of 2000 personas each, including a set of ‘realistic’ personas and a set of cinematic personas, showed that the correlation matrix among personality constructs was affected by the persona set. It is concluded that evaluating questionnaires and research hypotheses prior to engaging with real individuals holds promise.

As automated vehicles (AVs) become increasingly popular, the question arises as to how cyclists will interact with such vehicles. This study investigated (1) whether cyclists spontaneously notice if a vehicle is driverless, (2) how well they perform a driver-detection task when explicitly instructed, and (3) how they carry out these tasks. Using a Wizard-of-Oz method, 37 participants cycled a designated route and encountered an AV multiple times in two experimental sessions. In Session 1, participants cycled the route uninstructed, while in Session 2, they were instructed to verbally report whether they detected the presence or absence of a driver. Additionally, we recorded participants' gaze behaviour with eye-tracking and their responses in post-session interviews. The interviews revealed that 30% of the cyclists spontaneously mentioned the absence of a driver (Session 1), and when instructed (Session 2), they detected the absence and presence of the driver with 93% accuracy. The eye-tracking data showed that cyclists looked more frequently and for longer at the vehicle in Session 2 compared to Session 1. Additionally, participants exhibited intermittent sampling of the vehicle, and they looked at the area in front of the vehicle when it was far away and towards the windshield region when it was closer. The post-session interviews also indicated that participants were curious, but felt safe, and reported a need to receive information about the AV's driving state. In conclusion, cyclists can detect the absence of a driver in the AV, and this detection may influence their perception of safety. Further research is needed to explore these findings in real-world traffic conditions.

As global demand for fruits and vegetables continues to rise, the agricultural industry faces significant challenges in securing adequate labor. Robotic harvesting devices offer a promising solution to address this issue. Harvesting delicate fruits, such as blackberries, presents unique open challenges due to their fragility. This paper introduces BerryTwist, a prototype robotic gripper specifically designed for blackberry harvesting. The gripper features a fabric tube mechanism that uses motorized twisting action to gently envelop the fruit, ensuring uniform pressure application and minimizing damage. The twisting motion is transferred to the tube through a compliant mechanism, thus maintaining the overall softness of the structure. We thoroughly tested BerryTwist, paying particular attention to the effect of varying tube properties. We developed three types of tubes varying in elasticity and compressibility, using foam padding, spandex, and food-safe cotton cheesecloth. Performance testing focused on assessing each gripper's ability to detach and release blackberries, with an emphasis on quantifying damage rates. The results indicate that the proposed gripper achieved an 82% success rate in detaching blackberries and a 95% success rate in releasing them, demonstrating its promising potential for robotic harvesting applications. Finally, we will demonstrate the robotic harvesting operation by establishing a simple farm setup and integrating the gripper with Franka Emika's robot manipulator.

Within a year of its launch, ChatGPT has seen a surge in popularity. While many are drawn to its effectiveness and user-friendly interface, ChatGPT also introduces moral concerns, such as the temptation to present generated text as one’s own. This led us to theorize that personality traits such as Machiavellianism and sensation-seeking may be predictive of ChatGPT usage. We launched two online questionnaires with 2000 respondents each, in September 2023 and March 2024, respectively. In Questionnaire 1, 22% of respondents were students, and 54% were full-time employees; 32% indicated they used ChatGPT at least weekly. Analysis of our ChatGPT Acceptance Scale revealed two factors, Effectiveness and Concerns, which correlated positively and negatively, respectively, with ChatGPT use frequency. A specific aspect of Machiavellianism (manipulation tactics) was found to predict ChatGPT usage. Questionnaire 2 was a replication of Questionnaire 1, with 21% students and 54% full-time employees, of which 43% indicated using ChatGPT weekly. In Questionnaire 2, more extensive personality scales were used. We found a moderate correlation between Machiavellianism and ChatGPT usage (r = 0.22) and with an opportunistic attitude towards undisclosed use (r = 0.30), relationships that largely remained intact after controlling for gender, age, education level, and the respondents’ country. We conclude that covert use of ChatGPT is associated with darker personality traits, something that requires further attention.

Exterior vehicle sounds have been introduced in electric vehicles and as external human–machine interfaces for automated vehicles. While previous research has studied the effect of exterior vehicle sounds on detectability and acceptance, the present study takes on a different approach by examining the efficacy of such sounds in deterring people from crossing the road. An online study was conducted in which 226 participants were presented with different types of synthetic sounds, including sounds of a combustion engine, pure tones, combined tones, and beeps. Participants were presented with a scenario where a vehicle moved in a straight trajectory at a constant velocity of 30 km/h, without any accompanying visual information. Participants, acting as pedestrians, were asked to hold down a key when they felt safe to cross. After each trial, they assessed whether the vehicle sound was easy to notice, whether it gave enough information to realize that a vehicle was approaching, and whether the sound was annoying. The results showed that sounds of higher modeled perceived loudness, such as continuous tones with high frequency, were the most effective in deterring participants from crossing the road. The tested intermittent beeps resulted in lower crossing deterrence than continuous tones, presumably because no valuable information could be derived during the inter-pulse intervals. Tire noise proved to be effective in deterring participants from crossing while being the least annoying among the sounds tested. These results may prove insightful for the improvement of synthetic exterior vehicle sounds.

Raven matrices are widely considered a pure test of cognitive abilities. Previous research has examined the extent to which cognitive strategies are predictive of the number of correct responses to Raven items. This study examined whether response times can be explained directly from the centrality and visual complexity of the matrix cells (edge density and perceived complexity). A total of 159 participants completed a 12-item version of the Raven Advanced Progressive Matrices. In addition to item number (an index of item difficulty), the findings demonstrated a positive correlation between the visual complexity of Raven items and both the mean response time and the number of fixations on the matrix (a strong correlate of response time). Moreover, more centrally placed cells as well as more complex cells received more fixations. It is concluded that response times on Raven matrices are impacted by low-level stimulus attributes, namely, visual complexity and eccentricity.