robotic assistants - intechopen
TRANSCRIPT
Robotic Assistants: Science meets Fiction
Carme Torras @
Robots>10
Doctors28
PhD students37
Technicians10
Support staff14
People:
Institut de Robòtica i Informàtica Industrial
Perception and Manipulation Lab
Robot manipulators in human environments
✦ Learning from demonstration ✦ Planning and perception for manipulation
✦ Perception of rigid and non-rigid objects
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35
1. From industrial to assistive robotics
2. Research challenges: illustrative projects
3. Ethical and social implications
Institut de Robòtica i Informàtica Industrial
4
Presentation outline
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 5
Assistance to disabled and elderly people
Urban guidance, shopping helpers, cleaning
Co-workers in workshops and factories
Training and education
Industrial robots Assistive robots
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
6
World’s largest civilian robotics programme
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
7
Industrial robots Assistive robots
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Programmed by experts
Caged
Rigid objects -
Accurate
Fixed sequences
Non-interactive
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Challenges
8
Industrial robots
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Programmed by experts
Caged
Rigid objects -
Accurate
Fixed sequences
Non-interactive
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
9
Challenges Techniques
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Initial guidance + reinforcement
Modelling robot dynamics
Visual learning + task-oriented descriptors
Probabilistic state and action representations / uncertainty
Learning to plan
Learning from demonstrations
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35
1. From industrial to assistive robotics
2. Research challenges: illustrative projects
3. Ethical and social implications
Institut de Robòtica i Informàtica Industrial
10
Presentation outline
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
11
Challenges Techniques
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Initial guidance + reinforcement
Modelling robot dynamics
Visual learning + task-oriented descriptors
Probabilistic state and action representations / uncertainty
Learning to plan
Learning from demonstrations
Easy to program / safe for people
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
13
Initial guidance + reinforcement learning
Dynamic Movement Primitives (DMP):
Compact, rescalable, intuitive parametrization
IROS'14
Easy to program / safe for people
Easy to program - bimanual skills
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
16
Challenges Techniques
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Initial guidance + reinforcement
Modelling robot dynamics
Visual learning + task-oriented descriptors
Probabilistic state and action representations / uncertainty
Learning to plan
Learning from demonstrations
Carme Torras @ / 3617
Garment recognition and pose estimation
Usual approach:
Re-grasping to place garment in a standard configuration eases perception but is slow…
Our approach:
Involved computer vision and machine learning algorithms for informed (task-oriented) one-shot grasping
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
18
Visual recognition for informed grasping
Clothing dataset with labelled parts
ICRA’12 - IROS’13 - EAAI, 2014
(Bag-of-Words + SVM) for part location
Visual recognition for informed grasping
Perceiving and manipulating other deformable objects
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Initial guidance + reinforcement
Modelling robot dynamics
Visual learning + task-oriented descriptors
Probabilistic state and action representations / uncertainty
Learning to plan
Learning from demonstrations
21
Challenges Techniques
IntellAct: Intelligent observation and execution of Actions and manipulations
Probabilistic representations / Learning to plan
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
23
On-line learning Starts with no previous knowledge Initial demonstration --> rules Adapts to changes --> rule prob. updates
Active requests for teacher help Guide the teacher to minimize interaction Explain planning failures Sketch options: new actions, state changes
Learning to plan: RL + demonstrations
Two options (in red)
ICRA’14, Special Issue "AI and Robotics", 2015
Learning to plan: Teacher help
Learning to plan for goal-directed execution
IntellAct Online Perception
IntellAct Online Monitoring and Execution
Carme Torras @ / 35
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Easy to program by non-experts
Intrinsically safe for people
Able to perceive and manipulate deformable objects
Tolerant to noisy perceptions and inaccurate actions
Capable of goal-directed execution
Collaborating with people
Initial guidance + reinforcement
Modelling robot dynamics
Visual learning + task-oriented descriptors
Probabilistic state and action representations / uncertainty
Learning to plan
Learning from demonstrations
28
Challenges Techniques
Learning to collaborate from demonstrations
AAAI'13
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35
1. From industrial to assistive robotics
2. Research challenges: illustrative projects
3. Ethical and social implications
Institut de Robòtica i Informàtica Industrial
30
Presentation outline
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 31
Moral issues: military, legal liability, digital gap, …
Robotic assistants differ from other technologies in entering the domain of human feelings.
Ethical and social implications
How will human nature change with increasing H-R interaction?
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 32
Artificial retinas, sensorized dresses, exoskeletons, telepresence… robotic prostheses expand our body
Robots and humans… two types of ties
Living with butlers and artificial nannies, learning from robotic teachers, sharing work and leisure with humanoids... will enhance our intellectual and social habits? will develop new ones?
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 33
➢ N. Sharkey & A. Sharkey: "The crying shame of robot nannies: an ethical appraisal". (Pros and cons of robot nannies)
Classic science-fiction stories:
➢ I. Asimov (1950) "I, Robot" (protect vs. freedom) ➢ Ph.K. Dick (1955) "Nanny" (animate vs. inanimate) ➢ R. Bradbury (1969) "I sing the body electric" (acceptance-
immortality, sincerity)
Science meets fiction
2010: Special issue of the journal Interaction Studies
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 34
Can the effects of technology on human evolution be predicted?
Methodological difficulties: ● Appearance of unforeseen uses for devices (Ihde, 2004) ● Limitations of language to describe the future: “it is
through technique that we perceive the sea as navigable” (Heidegger)
● Cannot be studied separately from the socio-cultural context: Social construction of reality (Berger and Luckmann, 1966)
Joint work with F. Ballesté (Humanities, UOC)
CELEBRATING 10 YEARS 33
C
ELEBRATING Y E A R S
Carme Torras @ / 35 35
“It is the relationships that we have constructed which in turn shape us”
Robert C. Solomon, “The Passions”
Robotic assistants: social and ethical implications
Neal Stephenson, “Innovation starvation”
“Science fiction provides coherent scenarios of a technology integrated into a society, into an economy, and into people’s lives”
The future holds exciting technoscientific, ethical and anthropological challenges