XWiki - Foundation.Human Factors.Human-Agent Teamwork

1

In projects such as ours, where humans and agents cooperate and work as a team towards a common goal, the main concepts of teamwork still hold. Therefore, it is interesting to examine how the most prominent of them, the joint activities, the knowledge processes, and the task management, translate when having one or more non-human members on the team.

2

3

=== Joint Activities ===

4

When performing activities together, being able to predict what others will do and adjust one's own behavior is crucial for effective coordination. Additionally, there should be some common ground established. A shared mental model, meaning a representation of the world known to all members of the team, is crucial for communication.

5

In our scenario, our agent should expect that the PwD might have forgotten to eat and hydrate and should strongly encourage these actions. In case the patient has already eaten though, it should not persist and cause further annoyance, but ask some different questions instead. This aspect of the robot's behavior is mostly showcased in the [[//Design Patterns//>>https://xwiki.ewi.tudelft.nl/xwiki/wiki/sce2022group01/view/Main/Design%20Patterns/]] section. The mutual knowledge of the world and the shared mental model is represented and communicated to the agent through the ontology, better described in the [[//Ontology//>>https://xwiki.ewi.tudelft.nl/xwiki/wiki/sce2022group01/view/Main/Ontology/]] section.

6

7

=== Knowledge Processes ===

8

In terms of a team, there are different types of knowledge circulating. Task-related knowledge, meaning what are the tasks and who is in charge of them is one type, while team-related knowledge, meaning which members of the team have specific skills and who have information related to different tasks is another. There are also different team structures. Some teams have members that know more than others, some implement subteams responsible for their own subtasks, while some have their team members constantly change duties, where the spread of information is important to avoid any loss. In that case, a transaction memory system can be distributed across different team members so that each of the members has knowledge over their own expertise and also who in the team is overseeing each of the rest of the tasks.

9

The robot that we developed has both task- and team-related knowledge, knowing how to perform certain tasks but also who can give them the information that is needed to do so. Stakeholders responsible for the personalization of the robot, which can be the PwD, a relative, or a health care professional, provide the factual information needed for the robot to then behave on each own and interact with the patient.

10

11

=== Task Management ===

12

Regarding task management in a team, a principal aspect of it is coordination, the managing of actions from each member of the team to have as smooth a result as possible. In bottom-up coordination, a member follows the rest of the team, whereas, in top-down coordination, a manager would assign roles to the rest of the team. Since our robot's main goal is to encourage and make sure the PwD has eaten, we could say that the coordination follows more of a top-down approach. With the use of a robot for this task, we also aim at taking some of the workloads off of the caregiver, especially if it is a family member since conversations like this can often be taxing on familial relationships and bonds, as well as the relatives' lives which are usually affected greatly.

Wiki source code of Human-Agent Teamwork