Public Evaluation on Potential Social Impacts of Fully Autonomous Cybernetic Avatars for Physical Support in Daily-Life Environments: Large-Scale Demonstration and Survey at Avatar Land

However, to ensure the robustness of exophora resolution during the daily-life support CA R&D zone demonstration, we preloaded a user's pose skeleton and transcribed instruction data prepared offline, and manually recorded object locations as the model inputs. In addition, instead of demonstrating active exploration with visitors, which is time-consuming by nature, a pre-recorded video of the exploration performed on the site of Avatar Land was shown on TV displays. B. Object Manipulation After moving near the target object inferred by the ex-ophora resolution model, the Fetch CA picks up the object and places it on a Kachaka CA for delivery to the user, as the daily-life support CA R&D zone demonstration was designed to emphasize multi-CA collaboration. Although the position of the object is assumed to be known, in reality, there may be errors in self-position estimation, or the object position may have changed since the exophora resolution. Therefore, additional object detection was conducted with Detic [17] before grasping the target object for confirmation. The detected results were then combined with depth information to estimate the position of the target object. The position for placing the object on the Kachaka CA also needed to be determined similarly, so AR markers were attached to the shelf to compute the relative positions between the two CAs.