Section: New Results

Locomotion and Interactions between Walkers

MimeTIC is a leader in the study and modeling of walkers' visuo-motor strategies. This implies to understand how humans generate their walking trajectories within an environment. This year, one main focus was to consider how the interaction models change with specific populations (including kids, older adults, concussed athletes or person on a wheelchair) as well as in specific environment (including narrow sidewalk, or environment with varying social context).

Effect of Foot Stimulation on Locomotion

Participants : Anne-Hélène Olivier, Armel Crétual [contact] , Carole Puil.

Medio-Intern Element (EMI®) is a thin plantar insert used by podiatrists to treat postural deficiency. It was shown an influence of a 3 mm high EMI on Medio-Lateral (ML) displacement of the Centre of Pressure (CoP) of healthy participants in quasi-static standing. Recently it has been demonstrated that EMI has an impact on eyes vergence, and especially in population with plantar postural dysfunction. These effects were weakly assessed however and only using static tasks. Therefore, the objective of this work [53], [52], [41], was to evaluate the effect of the EMI while performing a locomotor task. We expected a controlateral deviation of the trajectory when this insert was located under one foot. Indeed, in previous studies dealing with bottom-up control of locomotion, it was shown that a 30 min podokinetic stimulation leads to a ML deviation of the trajectory when participants were asked to walk in a straight line with eyes closed. 20 healthy participants volunteered for this study. They participated into 3 different sessions in random order: either without EMI, with EMI under the right foot or under the left foot. Each session involved first, static tasks (with and without vision) to compare with previous work, then, dynamic locomotor tasks with 6 different conditions mixing trajectory (straight walking, 90° left or right turn) and vision (with and without vision) in random order. In static conditions, we computed the average ML position of the CoP. In dynamic conditions, we analyzed the difference in the final orientation of the locomotor trajectory with and without vision with an EMI with respect to this difference without the EMI. No significant effect of the EMI was observed for either static or dynamic conditions. Our results do not confirm the previous work in static conditions. Future work is needed to better understand the effect of this insert. In particular, our participants were healthy and it could be interesting to evaluate this effect in participants with postural deficiencies. These results would have an application in the design of new clinical tests.

Collision Avoidance between Walkers on a Curvilinear Path

Participants : Anne-Hélène Olivier [contact] , Armel Crétual, Richard Kulpa, Anthony Sorel.

Crowded public spaces require humans to interact with what the environment affords to regulate interpersonal distance to avoid collisions. In the case of rectilinear trajectories, the collision avoidance behaviours have been extensively studied. It has been shown that the perceived action-opportunities of the walkers might be afforded based on a future distance of closest approach (also coined ‘Minimal Predicted Distance’, MPD). However, typical daily interactions do not always follow rectilinear but also curvilinear trajectories. In that context, it has been shown that a ball following a curvilinear trajectory can be successfully intercepted. However, it remains unclear whether the collision avoidance strategies in the well-studied linear trajectories can be transferred to curvilinear trajectories. Therefore, the aim of this work [44] was to examine collision avoidance behaviours when interacting with walkers following curvilinear trajectories. An experiment was designed using virtual reality in which 22 participants navigated toward a goal in a virtual environment with a joystick. A Virtual Human (VH) crossed the path of the participant from left and right with varying risks of collision. The VH followed either a curvilinear path with a fixed radius of 5 m or 10 m, approaching from in-front of and behind the participant, or a control rectilinear path. The final crossing distance, the number of collisions and inversions of initial crossing order were analysed to determine the success of the task. Further, MPD evolution over time and specific timing events was analysed across conditions. For a curvilinear path with a 5 m radius there were significantly more collisions when the VH approached from behind the participant, and significantly more inversions of the initial crossing order when the VH approached from in-front than the control rectilinear path. Final crossing distance was shorter when the VH followed a path with a 5 m radius from behind the participant. Finally, the evolution of the MPD over time was similar for paths with a 10 m radius when compared to the control rectilinear path, whereas the 5 m curvilinear paths had significant differences during the interaction. Overall, with few collisions and few inversions of crossing order we can conclude that participants were capable of interacting with virtual walkers on curvilinear trajectories. Further, the task was solved with similar avoidance adaptations to those observed for rectilinear interactions. However, paths with a smaller radius had more reported collisions and inversions. Future work should address how a curved trajectory during collision avoidance is perceived.

Collision Avoidance in Person-Specific Populations

Participants : Anne-Hélène Olivier [contact] , Armel Crétual.

In the frame of the Inria BEAR associate team, we have used our 90 crossing paradigm to understand visuo-motor coordination in specific population. This is important, not only from a theoretical point of view but also to design more individual model of human locomotion in a dynamic environment.

Figure 6. Illustration of person-person interaction experiments in a) kids, b) older adults, c) previously concussed athletes, d) a person on an electric powered wheelchair

We first investigated the effect of age on visuo-motor coordination by considering a collision avoidance task in kids (8-12 years) and older adults (65-74 years) as illustrated on Figure 6a,b. On one hand, middle-aged children have been shown to have poor perception-action coupling during static and dynamic collision avoidance tasks. Research has yet to examine whether perception-action coupling deficits persist in a dynamic collision avoidance task involving a child and another walker. In this work [26], [54], we invesigated whether the metric MPD(t) be used to examine collision avoidance strategies between children and adults. To this end, eighteen children (age: 10 ± 1.5 years) and eighteen adults (34 ± 9.6 years) walked while avoiding another participant (child or adult). Groups of three children and three adults were recruited per session. The results demonstrated that (1) MPD(t) can be used to predict future collisions in children, (2) MPD(t) is an absolute measure that is consistently lower when a child is involved compared to two adult walkers, (3) the individual passing second, even when it is a child, contributes more to MPD(t) than the walker passing first. It then appears that children have developed adult-like strategies during a collision avoidance task involving two walkers. Body anthropometrics should be considered when determining collision avoidance strategies between children and adults. On the other hand, every year, 1 in 3 older adults are likely to fall at least once and many falls occurs while walking where an individual needs to adapt to environmental hazards. Studies with older adults interacting within an environment showed difficulties in estimating time to arrival of vehicles, larger critical ratio and more variability in door aperture task as well as larger clearance distance when avoiding a moving object. The current study [51] aims to identify whether differences in collision avoidance behaviours of older adults during a person-person collision avoidance task are the result of age-related visuomotor processing deficits. Results showed that no collision occurred, where older and younger adults were able to act appropriately. However, larger threshold were needed to trigger avoidance when an older adult is second in crossing order, possibly due to visuomotor delays. Moreover, we observed more crossing inversions with older adults, which may suggest a poor visumotor processing. Finally, the clearance distance was samller when to older adults interact with each others, resulting in “risky” behaviours. Interestingly, social factors seems to be involved since when a young and an older adults interact, the young adult contribute more to solve the collision avoidance task.

In close relation with the Application Domain "Sports", we also investigated visuo-motor coordination during locomotion in previously concussed rugby-players (Figure 6c). Despite adherence to return-to-play guidelines, athletes with previous concussion exhibit persistent visuomotor deficits during static balance and visuomotor integration tasks such as collision avoidance months after returning to sport. Previous research in collision avoidance was done in a static setting, however less is known about visuomotor strategies utilized in dynamic scenarios, such as person-person interactions. In this context, during a collision avoidance locomotor task, individuals make adjustments to their path and/or their velocity in response to a risk of collision. These adjustments ensure that the clearance distance would be large enough such that no collision occurs. However, athletes with previous concussion may demonstrate impaired performance during a collision avoidance task requiring path adjustments based on visual information. The purpose of this study [55] was to investigate collision avoidance strategies when avoiding another walker between previously concussed athletes and healthy athletes. We hypothesized that previously concussed athletes would demonstrate altered trajectory adaptation and changes in individual contribution to the avoidance compared to healthy athletes. Preliminary results show that individuals with previous concussion demonstrated trajectory adaptation behaviours consistent with healthy athletes and young adults. However, previously concussed athletes passed with a reduced distance between themselves and the other walker when they are second in passage order at the crossing point. Athletes who have sustained a previous concussion show decreased collision avoidance behaviour. This behaviour results in a higher risk of a collision occurring, as individuals showed reduced contributions (i.e. creating physical space) to the avoidance of the collision. This change in typical behaviour on a visuomotor task may indicate a persistent deficit in perceptual abilities following concussion. Although trajectory adaptations were consistent with healthy athletes, these results suggest that athletes with previous concussion remain at an elevated risk of collision and possible injury following concussion recovery. This study provides novel insights and additional evidence that visuomotor and perceptual impairments persist following return to play in previously concussed athletes. Additionally, this protocol has important implications for the assessment and rehabilitation of visuomotor processes that are affected following a concussion. Future research could further develop this protocol to be used in sideline assessment, and guide treatment of concussions past clinical recovery.

Collision Avoidance between a Walker and an Electric Powered Wheelchair: Towards Smart Wheelchair

Participants : Anne-Hélène Olivier [contact] , Armel Crétual.

In collaboration with Marie Babel and Julien Pettré from Inria Rainbow team, we are interested in the development of smart electric powered wheelchairs (EPW), which provide driver assistance. Developing smart assistance requires to better understand interactions between walkers and such vehicles. We focus on collision avoidance task between an EPW (fully operated by a human) and a walker, where the difference in the nature of the agents (weight, maximal speed, acceleration profiles) results into asymmetrical physical risk in case of a collision, for example due to the protection EPW provides to its driver, or the higher energy transferred to the walker during head-on collision. In this work [39], [47], our goal is to demonstrate that this physical risk asymmetry results into differences in the walker’s behavior during collision avoidance in comparison to human-human situations. 20 participants (15 walkers and 5 EPW drivers) volunteered to this study. The experiment was performed in a 30mx20m gymnasium. We designed a collision avoidance task, where an EPW and a human walker moved towards a goal with orthogonal crossing trajectories (Figure6d). We recorded their trajectory among 246 trials (each trial being 1 collision avoidance). We compared the predicted passage order when they can first see each other with the one observed at the crossing point to identify if inversions occur during the interaction. Note that during walker-walker interactions it was shown that the initial passage order is almost systematically preserved all along the interaction up to the crossing point. We also computed the shape-to-shape clearance distance. We observed 23.7% of passage order inversion, specifically in 20.8% of trials where walkers were supposed to cross first, they crossed second. This means that walkers were more likely to pass behind the EPW than in front. On average, human walkers crossed first when having sufficient advance on the wheelchair to reach the crossing point. We estimated this advance up to 0.91m. The shape-to-shape clearance distance was influenced by the passage order at the crossing point, with larger distance when the walker cross first (M=0.78m) than second (M=0.34m). Results show that walkers set more conservative strategies when interacting with an EPW. By passing more frequently behind the EPW, they avoid risks of collisions that would lead to high energy transfer. Also, when they pass in front, they significantly increase the clearance distance, compared to cases where they pass behind. These results can then be linked to the difference in the physical characteristics of the walkers and EPW where asymmetry in the physical risks raised by collisions influence the strategies performed by the walkers in comparison with a similar walker-walker situation. This gives interesting insights in the task of modeling such interactions, indicating that geometrical terms are not sufficient to explain behaviours, physical terms linked to collision momentum should also be considered.

Collision Avoidance on a Narrow Sidewalk

Participant : Anne-Hélène Olivier [contact] .

In the context of transportation research and a collaboration with the colleagues of Ifsttar (LEPSIS, LESCOT), we investigate person-person interaction when walking on a narrow sidewalk [34]. Narrow sidewalks are not the result of imagination nor a heritage of the former urban planning in the oldest cities. They exist in many modern cities, a simple web query provides a lot of examples in the world. In most cases, two pedestrians walking in opposite way cannot stay both on the sidewalk when they cross: one has to give a free way on the curb by stepping down on the road, which can generate risky situations for pedestrians. These situations are nowadays underestimated and so are the associated risk. In this context, driving simulators and walking simulators are useful tools to conduct studies in a safe environment with controlled conditions. Therefore, they can allow improving our knowledge on the way pedestrians interact on a narrow sidewalk and how drivers can react when facing this situation. This contribution aims to model the behaviours of simulated pedestrians, Non Player Characters (NPC). Using an interdisciplinary framework, we first identified from the literature psychosocial factors that should be involved in such interactions. Then, we designed a questionnaire to evaluate the impact of these factors on the perception of these interaction. Based on the main factors, we developed a perception model, and we modified the ORCA model, which is one of the most used for pedestrian collision avoidance simulation. Finally, we assessed the consistency of all our simulated interactions with a user study.

Shared Effort Model During Collision Avoidance

Participants : Anne-Hélène Olivier [contact] , Armel Crétual.

In collaboration with Jose Grimaldo da Silva and Thierry Fraichard (Inria Grenoble), we finally designed a shared-effort model during interaction between a moving robot and a human relying on walker-walker collision avoidance data. [33]. Recent works in the domain of Human-Robot Motion (HRM) attempted to plan collision avoidance behavior that accounts for cooperation between agents. Cooperative collision avoidance between humans and robots should be conducted under several factors such as speed, heading and also human attention and intention. Based on some of these factors, people decide their crossing order during collision avoidance. However, whenever situations arise in which the choice crossing order is not consistent for people, the robot is forced to account for the possibility that both agents will assume the same role, a decision detrimental to collision avoidance. In our work we evaluate the boundary that separates the decision to avoid collision as first or last crosser. Approximating the uncertainty around this boundary allows our collision avoidance strategy to address this problem based on the insight that the robot should plan its collision avoidance motion in such a way that, even if agents, at first, incorrectly choose the same crossing order, they would be able to unambiguously perceive their crossing order on their following collision avoidance action.