Probabilistic Robotics and Models of Gaze Control

Transcription

1 Probabilistic Robotics and Models of Gaze Control Dr. José Ignacio Núñez Varela MICCS 2015

2 Part I: Probabilistic Robotics

3 Imagen:

4 Imagen:

5 Why do we need robots?

6 Imagen:

7 Imagen:

8 Imagen:

9 Imagen:

10 Imagen:

11 Imagen:

12 Imagen:

13 But the kind of robot we really want is

14 Imagen: cb /bigbangtheory/images/e/ed/the-big-bang-theory-the-robotic- Manipulation.jpg

15 No, not really. What about this

16 Imagen:

17 Imagen:

18 Imagen:

19 Unfortunately, there are still many things to solve first

20 Basic model of robot interaction

21

22 Sensing Planning Acting Asimo Honda Picture credits:

23 We need intelligent robots!

24 Intelligent robots Learning Reasoning Decision-making Planning Understanding Common sense PR2 Willow Garage

25 Robots have to be able to accomodate the enormous uncertainty that exists in the physical world. Imagen:

26 But, what is uncertainty?

27 Being not certain about something. Imagen:

28 The estimated percentage by which an estimated or calculated value may differ from the true value.

29 Imagen:

30 Let s see an example Robot grasping

31

32 The robot might not have a good estimate of where the object is

33

34 What factors contribute to the robot's uncertainty?

35 Robot Environments Well structured environment << uncertainty Not structured environment >>> uncertainty Imagen:

36 Robot Sensors Sensors are limited in what they can perceive (e.g., physical limitations affect range and resolution) Sensors are subject to noise Sensors can break Imagen:

37 Robot Actuators Motors are, at some extent, unpredictable Control noise, wear-and-tear, mechanical failure Imagen:

38 Robot s Internal Models All internal models of the world are approximate Model errors have often being ignored Imagen:

39 Algorithmic Approximations Robots are real time systems, thus limiting the amount of computation being carried out Algorithms need to be approximated Imagen:

40 Robots are forced to act even though they don't have sufficient information to make decisions with absolute certainty. Imagen:

41 Managing uncertainty is possibly the most important step towards robust real-world robot systems. - Thrun, Burgard and Fox

42 Probabilistic Robotics Key idea: Represent uncertainty explicitly using the calculus of probability theory. Instead of relying on a single best guess, probabilistic algorithms represent information by probability distributions over a whole space of guesses.

43 Mobile Robot Localization Imagen:

44 Mobile Robot Localization The map is given The robot wants to know where it is

45 Mobile Robot Localization The robot assumes a uniform probability distribution of where it is (it is likely to be in any place in the map)

46 Mobile Robot Localization The robot s belief increases after sensing a door (data is integrated into the old belief)

47 Mobile Robot Localization The robot moves some distance (Its belief moves as well, but the movement introduces some noise)

48 Mobile Robot Localization The robot senses a door once again, and its belief of where it is increases

49 Bayes Theorem Imagen:

50 Bayes Theorem Prior probability

51 Bayes Theorem Data

52 Bayes Theorem Posterior probability

53 Part II: Gaze Control Imagen:

54 Gaze Control Biological perspective Machine perspective Jason Babcock icub.org

55 Why study gaze control?

56

57

58

59

60 Foveal Vision cellfield.ca Michael Land

61

62

63 Eye Movements Saccades Aim: Shift the fovea to obtain high resolution samples Rapid jump-like movements (900 /sec) Ballistic (trajectory cannot change) Stereotyped (follow the same pattern) Voluntary and involuntary

64 Saccade Sequence

65 We perform hundreds or even thousands of saccades every day!

66 How does the brain decide where to fixate next?

67 Active Vision Ilya Repin

68 Yarbus

69 Task and context determine where to fixate next

70 Vision and Action Mary Hayhoe

71 Uncertainty Reduction

72 What mechanisms a rational decision maker could employ to select a gaze location optimally, or near optimally, given limited information and limited computation time during the performance of a task? Engineering science goal How humans select the next gaze location? Human behavioural goal

73 Gaze Control Processes

74 icub Humanoid Robot icub.org

75 Two problems where to look gaze allocation

76 Pick & Place Task

77

78

79

80

81 Models of Gaze Control Based on uncertainty reduction (Uncertainty) Based on rewards and uncertainty (Rew+Unc) Based on rewards, uncertainty and gain (Rew+Unc+Gain)

82 One-step look ahead gaze control What would happen if I look at entity e i?

83 Uncertainty Reduction How much uncertainty is reduced if I look at entity e i? X

84 Reward and Uncertainty How much value am I expected to get after looking at entity e i? X

85 Reward, Uncertainty & Gain Which motor system would get more benefit if gaze is allocated to it? X

86

87 Conclusions We need robots! There is still much to do before we can buy our assistant robot You can contribute to make this happen!

88 Thank You!! Botodesigns / Chen Reichert jose.nunez@uaslp.mx Website:

89 Reach/Grasp Sensitivity

90 Observation Noise

91 Field of View