Lecture 14 - Intention Reading and Context Awareness

Question 1

When do you speak about joint attention?

Answer 1

When two people are paying attention to the same thing

Question 2

What are the three forms of joint attention?

Answer 2

eye contact
object
workspace

Question 3

Co-operating agents must (3 parts):

Answer 3

Incorporate knowledge about the co-actor(s)
through action observation
ahead of time

Question 4

What are the four (multiple) levels of co-ordination?

Answer 4

Kinetic level (forces, timing?)
Action level (what to do?)
Goal level (what for?)
Reasoning (how to?)

Question 5

How to infer the action goal of the other agent?

Answer 5

By using simulation theory: use own motor system to simulate actions of others

Examples:

• Motor control theory
  
  • foward modelling: predict consequences of actions
  • action observations: predict observed action from action repertoire
• Robotics
  
  • direct mapping of observed joint angles on those of own action repertoire

Question 6

What is the problem of joint action?

Answer 6

to infer the action of another agent, you need to have similar effectors and kinematics
perception depends on viewpoint

Question 7

Four components of the robot playing air-hockey

Answer 7

1. primitive recognition
2. primitive selectiion
3. sub-goal generation
4. action generation

Question 8

How to implement multiple movement?

Answer 8

1. Action primitives
   - pick up full teapot
   - pick up empty teapot
2. separate controllers
   - for each primitve
   - run in parallel

the best prediction wins

Question 9

Action production (how to)

Answer 9

• assign weights to each controller

- motor command is weighted average

Question 10

action observation

Answer 10

simulation theory: simulate actions

• generate predictions
• compare with observations

Question 11

Evidence for goal inference:

Answer 11

1. imitate goals rather than the effector (14 month old infants with imitation)
2. mirror neurons

Question 12

Two fundamental processes in Bayesian model of action goal inference:

Answer 12

• Turn likelihood into beliefs (Bayes’ rule)

- belief propagation (law of total probability)

Question 13

Bayes rule combines … (2)

Answer 13

evidence from observation

prior knowledge

Question 14

Law of total probability …

Answer 14

propagates believes to more abstract levels

Question 15

Simulation theory …

Answer 15

use your ow action system for observations

Question 16

Goal inference

Answer 16

• decisions at the highest level

- generalize across movements and bodies

Question 17

How do humans interact? (4)?

Answer 17

• provide multiple consistent cues
• reduce noise by integrating cues
• infer goals rather than means
• combine prior knowledge with evidence from observation