CongreG8

About

CongreG8 dataset contains 380 human approaches groups trials and around 38 robot approaches groups trials. Each trial includes full-body motion capture data of all players and the robot (if exists) during a time period of 2-6 seconds with a frame rate of 120 fps. The data are exported as Comma Separated Values (CSV) files. Each CSV file contains 3D positions of all corrected markers and 3D position and rotation (in quaternion) of reconstructed skeletons or rigid bodies (the robot). The data are post-processed in order to manually correct mislabeled markers and replenish missing markers. The CongreG8 dataset also supports FBX format that is used in 3D animation systems and motion study applications. The dataset will support other data export formats such as C3D and BVH in the future. The research is approved by the research ethics board and has obtained IRB approval from the university IRB for public release of our dataset.

Dataset

You can find the samples of the dataset from this link:

https://www.dropbox.com/s/5q2i6n2lu4843m6/sample%20data.zip?dl=0

Scenario

To evoke interactions with a free-standing conversational group, we used a variant of a game called Who's the Spy which involves three group players and one observer. In every game round, each group player is given a card with a word on it. All cards are the same except one. The player who has the odd card is the spy. Unlike the traditional Who's the Spy game, in which group players vote to decide the spy, in our scenario the observer identifies the spy.

When the game starts, the group players take turns to describe objective properties of the word at hand. For example, if player A has a word Apple, he/she could say It's a fruit rather than I don't like to eat it. In the meanwhile, the observer stands far outside of the conversational group and listens to the ongoing conversation. Once the observer establishes the identity of the spy, he/she approaches and joins the conversational group in order to inform the other players.

As a part of our data collection efforts, we also collected robot-group interaction data using a physical Pepper robot. The robot plays the role of the observer and is remotely controlled using a Wizard of Oz (WoZ) approach. Note that the group players do not know that WoZ is used to control the robot while they are playing.

Contributors

Fangkai Yang, Royal Institute of Technology

Yuan Gao, Uppsala Univeristy

Ruiyang Ma, Royal Institute of Technology

Sahba Zojaji, Royal Institute of Technology

Christopher Edward Peters, Royal Institute of Technology

Contact

Fangkai Yang, Royal Institute of Technology