Distance Estimation in a Telepresence Scenario Using a 360° Monoscopic Camera, an AGV and an HMD

Abstract: As telepresence has an increasing number of use cases where veridical perception of distances is crucial, understanding influences on distance perception gets more important. Many researchers focus on stereoscopic virtual environments and most often find that distances get underestimated. But when an operator is remotely controlling a robot, the remote location is often monitored by a monoscopic camera, featuring less cues for distance estimation. Hence, this paper proposes the design and evaluation of a study on distance perception in a monoscopic telepresence environment. Live footage of a 360° camera on an autonomous guided vehicle was shown to the participants via head-mounted-display, while they controlled the movement of the vehicle with a controller. Participants (n = 16) operated the vehicle and completed three different tasks related to distance and spatial perception: (1) In the move-to-target task, distances to a target object were estimated by moving the robot to the target; (2) in the verbal estimation task, they verbally judged different distances; and (3) in the passability judgment task, they assessed whether the vehicle could fit through a gap between two objects. Tasks (1) and (2) were studied with 5 target distance values between 0.7 m and 4 m and task (3) was repeated for 5 gap widths that were up to 15 cm smaller or higher than the AGV width. Each estimation task was tested with and without a visual distance estimation aid (DEA), which consisted of a grid of known size that was displayed in the camera video. Statistical analysis suggests an influence of target distance. For the move-to-target and passability judgement task significant differences were found across all pairwise comparisons, but for verbal estimation only for distances in personal space, i.e. < 2 m. The applied distance estimation assistant did not have a significant impact on the results, improving move-to-target estimations slightly (6.6 %) while somewhat decreasing the accuracy of verbal estimations, i.e. raising overestimation by around 3 %. Besides the small impact, most participants reported feeling safer in correctly estimating distances with the DEA activated, while some reported higher strain on concentration or misjudging the positioning of the grid.Overall, distances were overestimated in the move-to-target task (M = 110.7 %, SD = 17.8 %), while verbal estimates were more balanced, with both over- and underestimations occurring (M = 99.3 %, SD = 22.2 %). In the passability judgment task, participants consistently overestimated the width of gaps, with up to 62.5 % judging them to be wider than they actually were. Underestimation occurred in only about 4 % of trials. This overall overestimation of distances does not fit to typical findings of distance perception in stereoscopic virtual environments and could be influenced by different factors, e.g. bad image quality of the camera or less spatial cues. It could also be attributed to the small camera height of 1 m, since empirical results in the literature unequivocally suggest that lowering the eye-height leads to increasing distance overestimation.

Keywords: Distance estimation, Telepresence, 360° monoscopic view, AGV, Distance Estimation Assistant

DOI: 10.54941/ahfe1007084

Cite this paper:

​