Movement data set for trust assessment (Drapebot robot cell/DLR)

In the Drapebot project, a worker collaborates with a large industrial manipulator in two tasks: collaborative transport of carbon fibre patches and collaborative draping. To realize data-driven trust assessement, the worker is equipped with a motion tracking suit and the body movement data is labeled with the trust scores from two standard Trust questionnaire (1. Trust perception scale - HRI, Schaefer 2016; 2. Trust in industrial human robo collaboration, Charalambous, et.al. 2016).

For this data set, data has been collected for the draping task from 21 participants all familiar with working with large industrial manipulators. For all sessions, body tracking was performed using the Xsens MVN Awinda tracking suit. It consists of a tight-fitting shirt, gloves, headband, and a series of straps used to attach 17 IMUs to the participant. After calibration the system uses inverse kinematics to track and log the movements of the participant at a rate of 60 Hz. The measurements include linear and angular speed, velocity, and acceleration of every skeleton tracking point (see XSENS manual for a detailed description of avaiable measurements).

Data organization

There are 21 files for 21 participants. The name of the files is PID01, where the number 01 is the participant. Each file contains all the data that was generated from the XSENS motion capture system. The files are xlsx files and for each sheet inside the excel file there are different types of data:

Segment Orientation - Quat

Segment Orientation - Euler

Segment Position

Segment Velocity

Segment Acceleration

Segment Angular Velocity

Segment Angular Acceleration

Joint Angles ZXY

Joint Angles XZY

Ergonomic Joint Angles ZXY

Ergonomic Joint Angles XZY

Center of Mass

Sensor Free Acceleration

Sensor Magnetic Field

Sensor Orientation - Quat

Sensor Orientation - Euler

See also: https://base.movella.com/s/article/Output-Parameters-in-MVN-1611927767477?language=en_US

For more information on each specific data and/or sensors please see the xsens manual (Link above)

Data Annotation

In each .xlsx file the first tab (sheet) is called "Markers". It annotates the starting frame of the individual tasks. The annotations are pickup, draping, return and some files may contain a also a "fail" annotation. Failed attempts should not be taken into consideration for model training.

The file trustscores.xlsx includes the results of the trust questionaires for each participant (scores for the individual items as well as the calculated overall trust scores).

Items for Trust perception scale - HRI, Schaefer 2016:

Which % of time does the robot

Function successfully

Act consistently

Communicate with people

Provide feedback

Malfunction

Follow directions

Meet the needs of the mission

Perform exactly as instructed

Have errors

Which % of the time is the robot:

Unresponsive

Dependable

Reliable

Predictable

Items for Trust in industrial human robo collaboration, Charalambous, et.al. 2016:

The way the robot moved made me uncomfortable

I felt I could rely on the robot to do what it was supposed to do  

The speed at which the gripper picked up and released the components made me uneasy    

I felt safe interacting with the robot

I knew the gripper would not drop the components

The size of the robot did not intimidate me

The robot gripper did not look reliable

I was comfortable the robot would not hurt me

I trusted that the robot was safe to cooperate with

The gripper seemed like it could be trusted

K. E. Schaefer, Measuring Trust in Human Robot Interactions: Development of the “Trust Perception Scale-HRI”. Boston, MA: Springer US, 2016, pp. 191–218.

G. Charalambous, S. Fletcher, and P. Webb, “The development of a scale to evaluate trust in industrial human-robot collaboration,” International Journal of Social Robotics, vol. 8, pp. 193–209, 2016.

在Drapebot项目中，操作人员与大型工业机械臂（large industrial manipulator）配合完成两项任务：碳纤维贴片协同转运与协同铺覆。为实现数据驱动的信任评估，操作人员配备动作捕捉服（motion tracking suit），其身体运动数据将通过两份标准信任问卷的评分进行标注：1. 人机交互信任感知量表（Trust perception scale - HRI, Schaefer 2016）；2. 工业人机协作信任量表（Trust in industrial human-robot collaboration, Charalambous et al. 2016）。

本数据集针对铺覆任务，共收集了21名熟悉大型工业机械臂操作的受试者的数据。所有测试环节均采用Xsens MVN Awinda动作捕捉系统进行身体追踪。该系统由紧身上衣、手套、头带以及用于将17个惯性测量单元（Inertial Measurement Unit, IMU）固定在受试者身上的多条绑带组成。经校准后，系统通过逆运动学（inverse kinematics）以60Hz的频率追踪并记录受试者的运动数据。所采集的测量数据包含所有骨骼追踪点的线速度、角速度以及加速度（可测量项的详细说明请参见XSENS用户手册）。

### 数据组织
本数据集包含21个文件，对应21名受试者，文件命名格式为PIDXX，其中XX为受试者编号（如PID01代表第1名受试者）。每个文件均包含XSENS动作捕捉系统生成的全部数据，文件格式为XLSX，每个Excel文件内包含多个工作表，分别存储不同类型的数据：
- 段朝向-四元数（Quat）
- 段朝向-欧拉角（Euler）
- 段位置
- 段速度
- 段加速度
- 段角速度
- 段角加速度
- ZXY顺序关节角
- XZY顺序关节角
- ZXY顺序人体工程学关节角
- XZY顺序人体工程学关节角
- 质心
- 传感器自由加速度
- 传感器磁场
- 传感器朝向-四元数（Quat）
- 传感器朝向-欧拉角（Euler）

参考链接：https://base.movella.com/s/article/Output-Parameters-in-MVN-1611927767477?language=en_US
如需了解各数据项及传感器的详细信息，请参阅XSENS用户手册（即上述链接）。

### 数据标注
每个XLSX文件的第一个工作表名为"Markers"，用于标注各任务的起始帧。标注类型包括拾取（pickup）、铺覆（draping）、返回（return），部分文件还会包含"失败（fail）"标注。模型训练时不应将失败的尝试纳入考量。

trustscores.xlsx文件包含每名受试者的信任问卷结果，涵盖各题项得分以及计算得到的总体信任评分。

#### Schaefer 2016年提出的人机交互信任感知量表（Trust perception scale - HRI）题项：
1. 机器人完成以下操作的时间占比为多少：
- 顺利执行任务
- 动作保持一致
- 与人员进行交互
- 提供反馈
- 发生故障
- 遵循指令
- 满足任务需求
- 严格按照指令执行
- 出现失误
2. 机器人在以下方面的表现占比为多少：
- 反应迟钝
- 可靠
- 可信
- 可预测

#### Charalambous等人2016年提出的工业人机协作信任量表题项：
- 机器人的移动方式令我感到不适
- 我认为可以信赖机器人完成其既定任务
- 夹爪拾取和释放部件的速度令我感到不安
- 与机器人交互时我感到安全
- 我确信夹爪不会掉落部件
- 机器人的体型并未让我感到畏惧
- 机器人夹爪看起来不可靠
- 我确信机器人不会伤害我
- 我信任该机器人，可与之安全协作
- 夹爪看起来值得信赖

### 参考文献
1. K. E. Schaefer, Measuring Trust in Human Robot Interactions: Development of the "Trust Perception Scale-HRI", Boston, MA: Springer US, 2016, pp. 191–218.
2. G. Charalambous, S. Fletcher, and P. Webb, "The development of a scale to evaluate trust in industrial human-robot collaboration", International Journal of Social Robotics, vol. 8, pp. 193–209, 2016.