Transcript Morphological computation

Morphological Computation –
Connecting Brain, Body, and Environment
2015-11-20
presenter : Kyung-Wha Park
Dong-Hyun Kwak
Biointelligence Laboratory
Seoul National University
Introduction What is morphological computation?

Morphology 형태학



Artificial intelligence and Biorobotics



생물의 형태, 구조, 발생 따위를 연구하는 학문.
Hardware in robotics : robot arms, actuators etc
Work out the principles underlying intelligent behavior
Enable us to
 Understand natural forms of intelligence
 Design and build intelligent systems
Traditionally..



Only focus on the control or the neural system, the brain, itself
Given morphology and then programmed, to do certain tasks.
Separation between hardware and software.
2
Introduction What is morphological computation?

Centered Notion : “Embodiment”




A complete organism that interacts with the real world
The idea that intelligence always requires a body?
Relation between physical and information (neural, control) process
Morphological computation

When morphology or materials take over some of the functions in
brain
Morphology
embodiment
Brain
(the control)
Real world
3
Introduction What is morphological computation?

Information self-structuring


Proper morphology and exploitation of intrinsic dynamics,
facilitate the induction of information structure
Information structure induced in the sensory data,
facilitating perception and learning

It can’t be understood by looking at internal brain only.

How the brain is embedded into the physical system?
How the organism interacts with the real world?
About connecting body, brain and environment.

© 2012-2014, SNU CSE Biointelligence Lab., http://bi.snu.ac.kr
4
Case Studies
Exploitation of Sensory Morphology

Eyebot:





Non-homogeneous arrangement of the facets compensates
Motion parallax facilitating neural processing.
Objects on the side travel faster across the visual field
“morphological computation”
Information self-structuring :
 Sensory stimulation is only induced, moving in a particular way
Artificial evolution on real robots
 Proper morphological arrangement
 Keeping lateral distance to an obstacle can be solved by proper
morphological arrangement : Frontally more dense than laterally
 Without changing anything inside the neural controller
5
Case Studies
Exploitation of Morphological and Material Properties

Cheap Grasping






How morphology, materials, and control interact to achieve grasping
18 DOF, tendon driven robotic prosthetic hand, deformable finger tips
For grasping, a simple control scheme, a “close” is applied
Because of the morphology of the hand, the elastic tendons, and the
deformable finger tips, the hand will automatically self-adapt to the object it
is grasping
The shape adaptation is taken over by the morphology of the hand, the
elasticity of the tendons, and the deformability of the finger tips, as the hand
interacts with the shape of the object
Efficient control, less Brain load
6
Case Studies
Exploitation of Morphological and Material Properties

Physical Dynamics



Swinging arm and hand movement :
controlling looks complex, but it’s simple
Proper morphology and exploitation of intrinsic dynamics,
facilitate the induction of information structure
Information Self-structuring



Proper morphology and exploitation of intrinsic dynamics, facilitate the induction of
information structure
Cross-modal associations can be formed between the visual, the haptic, etc
If the hand is swinging from the right side towards the center..

through sensory-motor coordinated behaviors such as seeing, reaching,
and grasping, sensory stimulation is not only induced but it also tends to
contain information structure
7
Case Studies
Managing Complex Bodies:
Adaptation through Morphological Changes

The Fin Ray Effect:




Two ribs interconnected by elastic connective tissue
Bends towards the direction of an external force
Provide an equal distribution of pressure in a dynamic system
“Aqua ray robot” by Festo [21].
 Resembles a manta ray
 Has a minimum setup with only three motors
 Although each motor is just driven back and forth,
 With “fin ray effect” provides the “Aqua ray” robot with a perfect and
cheap way to bend the wings and exploit the fluid dynamics
Brain (the control) + morphology
8
Case Studies
Managing Complex Bodies:
Adaptation through Morphological Changes

Exploiting Self-adaptation:
Cockroaches Climbing over Obstacles:





Cockroaches cannot move fast over uneven terrain,
but they can negotiate obstacles that exceed their body height
Complex bodies. one thoracic ganglion at each segment. 250 neurons
Face obstacle
 The brain, rather than changing the movement pattern
by evoking a different neural circuit, “re-configures” the shoulder joint
 Appropriately altered joint morphology
 But local neural circuits continue doing the same thing
“Brain outsourcing”
Advantages of controlling through morphological computation
 Can be solved efficiently with few neurons
 Freeing it from unnecessary control tasks
 Inherent stability of the local feedback circuits
 Designing controllers by exploiting mechanical change and feedback
9
Exploitation of Dynamics



Inspired from passive dynamic walkers, we will show 3
kinds of robots-”Stumpy, Puppy, Crazy Bird”.
And through these, we can know how materials and
dynamics are applied into robots.
Passive Dynamic Walkers



Without actuator nor motor, this can walk down the incline.
This kind of walking is very energy efficient – the energy is
supplied by gravity only.
Ecological niche (the environment in which the robot is
capable of operating) is extremely narrow: it only consists of
inclines of certain angles.
https://youtu.be/AdK7opXqoro?t=23s
10
Exploitation of Dynamics

Stumpy




The walking and hopping
robot.
압축공기를 이용한 actuator
가 장착되어, 평지에서도 걸
을 수 있게 디자인 됨.
Lower body is made of an
inverted “T” mounted on
wide springy feet.
The upper body is an upright
“T” connected to the lower
body by a rotary joint, the
“waist” joint.
11
Exploitation of Dynamics

Puppy


Fast and robust locomotion can be achieved with no sensory
feedback.(Feedback response times are too slow.)
Each leg has two standard servomotors and one springy
passive joint as a very simple kind of artificial “muscle”
12
Exploitation of Dynamics

Crazy Bird




To investigate how to gain controllability and increase the
behavioral diversity of robots.
Two motors running forward with constant speed.
Due to the robot's body shape and the interaction with the environ
ment behavioral diversity can emerge.
레고 마인드스톰 NXT kit로 만들어짐.
https://www.youtube.com/watch?v=vQIcbWbQug4
13
Exploitation of System Environment
Interaction

Wanda



The artificial fish, “Wanda” shows how the interaction with the
environment can be exploited in interesting ways
Achievement of moving to any position in 3D space with only one
degree-of-freedom (DOF) of actuation.
탄력있는 물질로 이루어진 꼬리 지느러미만을 통해 상하좌우로 움
직일 수 있다.
https://youtu.be/1FyF6Bac1fo?t=55s
14
Discussion

The integration scheme of these fundamental
principles and ideas.
https://www.youtube.com/watch?v=AdK7opXqoro
15
Discussion

질문 : 이러한 morphological computation 이 로봇 지능에
도 적용될 수 있는가?

답 : 기본적으로 앞서 살펴본 case study들은 움직임에 대
한 morphology의 예시이나, 이를 확장하여 로봇 지능에 인
간의 뇌 구조와 기작을 모방할 수 있다. 그러나 이를 위해
서는 충분한 인간의 뇌에 대한 이해가 필요하다.

질문 : 가장 핵심적으로 배워야 할 점?

답 : Embodiment가 반드시 real world 상에 있어야만 환경
과의 상호작용이 가능하고, 이로 부터 다양한 창발성이 일
어날 수 있다.
16