Benchmark

We seek to build a benchmark, an illustrative example that would serve as a common basis for discussion. Below, you will find a first proposal regarding this goal. It needs to be improved and discussed (e.g. letting room for norms definition or social context). However, we think it is crucial, to capture the subtlety of the work of each other and to built a common ground regarding the subject we want to tackle.

Feel free to contact us: fja@sciencesconf.org if you want to propose any improvement or modification. Thanks.

An illustrative example

Please feel free to play with this example since discussion is really welcome. However, it would be very useful if you clearly indicate in what way your work is helpful to deal with such an example, i.e. which "brick" your work could contribute to the framework.

A human and a robot have the goal to build a pile with 4 cubes and put a triangle at the top. One after the other, they should stack bricks in the expected order.Each agent has a number of cubes accessible in front of him and would participate to the task by placing its cubes on the pile. At the end, one of the agent should place a triangle at the top of the pile.

The initial state is the following:

 Untitled_1.jpg

Actions available for each agent are the following (with object = cube or triangle):

  • take an object on the table
  • take an object from the pile
  • put an object on the pile
  • give an object to the other agent
  • support the pile

Each agent is able to infer the state of the world so it knows:

  • where each object is
  • if an object is reachable for itself
  • if an object is reachable for the other one

Moreover, we assume each agent is able to observe the activity of the other. The expected final state could be one of the following:

final state 1final state 2

 

Possible deviations could be that:

  • an agent drops a brick on its side / in the opposite side
    • e.g. if the brick falls down on the opposite side so that it becomes unreachable for the intended agent to put it on the pile, consider whether the other agent should put the brick directly on the pile or give it to the intended agent
    • the pile collapses
    • ...

 

Moreover, during the execution of the task, a number of behaviors can arise, among all:

  • Proactive behavior:
    • one agent could [be lead to] help the other one by supporting the pile while the other places a brick on it
    • "Inactive" behavior:
      • one agent does not act at all
      • "Incorrect" behavior:
        • one agent does not pile bricks in the correct order
        • one agent removes a correctly placed brick from the pile

 

Finally, a negotiation phase should arise at the end to decide who put the triangle on the pile.

 

Scenario enhancement propositions:

(please to do not hesitate to do other propositions)

1) Have more variability in the task:

There should be more places where a choice is needed between the human or the robot to do the task. We could think that at the beginning each agent have all bricks available (so we would have two bricks 1, 2, 3 and 4, one for each for example)

2) Introduce social aspect in the task:

We could think to have several human "types" given their age or their (previous) knowledge.

 

Topics

We are seeking to frame joint action, interesting topics include (but are not restricted to):

  • joint goal establishment and negotiation
  • planning when the goal is a joint goal
  • joint goal / joint action description and modeling
  • joint goal / joint action execution, monitoring, turn-taking and timing management
  • agents world modeling and management (incl. theory of mind and shared knowledge)
  • agents commitment
  • what kind of communication during joint goal achievement and for what purpose
Online user: 1