Robotics 5
Transcript of Robotics 5
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 1/50
1
EXPERT SYSTEMS AND SOLUTIONS
Email: [email protected]
Cell: 9952749533www.researchprojects.info
PAIYANOOR, OMR, CHENNAI
Call For Research Projects Final
year students of B.E in EEE, ECE, EI,
M.E (Power Systems), M.E (Applied
Electronics), M.E (Power Electronics)
Ph.D Electrical and Electronics.
Students can assemble their hardware in our
Research labs. Experts will be guiding the projects.
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 2/50
2
Multiagent Control ofModular Self-Reconfigurable
Robots
Tad HoggHP Labs
Hristo BojinovJeremy Kubica
Arancha Casal
PARC·s modular robotics group
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 3/50
3
topics� modular robots
� multi-agent control� results
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 4/50
4
modular robots� collections of modules
² each module is a robot
� self-reconfigurable² modules can change connections
² so overall robot changes shape
� ´modular self-reconfigurableµ robots² MSR
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 5/50
5
why change shape?
� adjust shape to task
� e.g., locomotion² wheel, spider, snake, «
� e.g., manipulation² match ´fingerµ size to object size
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 6/50
6
topics� modular robots
² Proteo
² Prismatic
² future possibilities
� multi-agent control
� results
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 7/50
7
Proteo� rhombic dodecahedron
� space filling
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 8/50
8
Proteo� modules move over neighbors
each edge of cube is a diagonal of RD face
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 9/50
9
topics� modular robots
² Proteo
² Prismatic
² future possibilities
� multi-agent control
� results
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 10/50
10
prismatic MSR robots� modules connect via arms
� extending arms moves neighbors� examples² Crystalline robot (Dartmouth)
� moves in 2 dimensions
² TeleCube (PARC)� moves in 3 dimensions
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 11/50
11
TeleCube� cubes
� 6 independent arms
� 2:1 length ratio
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 12/50
12
neighbors cooperate to move
expandcontract
physical movevirtual move
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 13/50
13
topics� modular robots
² Proteo
² Prismatic
² future possibilities
� multi-agent control
� results
actual modules
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 14/50
14
devices for ´smart matterµ� micro-electomechanical (MEMS)
� bacteria
� molecular
� quantum
sensor + computer + actuatorsensor + computer + actuator
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 15/50
15
micromachines (MEMS)� made with photolithography
² e.g., programmable force fields (open loop)
� hard to assemble
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 16/50
16
biological machines� biotechnology: program bacteria
² e.g., T. Knight, R. Weiss at MIT AI Lab
� limited abilities
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 17/50
17
programs for bacteria� gene regulatory networks
� engineered changes give someprogram control over behavior
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 18/50
18
molecular machines� ribosomes:
² make proteins in cells
� protein motors² move material in cells
² ATP synthase rotor� size: 10nm
DNA mRNA
protein
See Nature, 386, 299 (1997)
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 19/50
19
molecular machines� carbon nanotubes and buckyballs
² strong, light, flexible, electronic devices
² easy to make
² hard to arrange
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 20/50
20
molecular machines� complex molecules for robot parts
� currently:² only theory
² hard to make
² hard to assemble
� potential: cheap, fast, strong parts
example medical applications:
R. Freitas, Jr., Nanomedicine, 1999
example designs:
E. Drexler , R. Merkle, A. Globus
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 21/50
21
quantum computers� potential: much faster algorithms
² e.g., factoring
� very difficult to build
amplitudes while solving a
10-variable 3-SAT instance
with 3 solutions
quantum search heuristic
Java demo: www.hpl.hp.com/shl/projects/quantum/demo
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 22/50
22
quantum machines� potential: detail control over materials
² e.g., interfere two ways to absorb light =>transparent
� very difficult to build
See T. Hogg and G. Chase, Quantum smart matter , 1996www.arxiv.org/abs/quant-ph/9611021
S. Lloyd and L. Viola, Control of open quantum systems dynamics, 2000
www.arxiv.org/abs/quant-ph/0008101
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 23/50
23
quantum machines� example: coin weighing puzzle
² quantum sensor finds bad coin in single try
See B. Terhal, J. Smolin, Single quantum querying of a database, 1997
www.arxiv.org/abs/quant-ph/9705041
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 24/50
24
devices: summary� smaller devices
² harder to make
² harder to connect, assemble
² greater potential capability
� but need many, cheap devices
� statistical or systems view
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 25/50
25
challenge: How
to build?� physical/engineering constraints
² unreliable parts
² misconnected
� limits early technology
� economics² build cheaply
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 26/50
26
challenge: How
to use?� information/computational constraints
² limited, changing info from environment
² computational complexity� e.g., planning optimal device use
� limits even mature technology
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 27/50
27
topics� modular robots
� multi-agent control
� results
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 28/50
28
control challenge� coordinate many modules
� sensor & actuator errors� decompose programming task to only need
² local info (small scale)
² high-level task description (large scale)
� e.g., grasp object of unspecified shape
cf., H. Simon: nearly decomposable systems
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 29/50
29
control before hardw
are?� many, small modules don·t yet exist
² hence, hardware details unknown
² but can study general issues
� control may simplify hardware design² e.g., manage in spite of defects
² identify compute/communicate tradeoffs
sensor + computer + actuatorsensor + computer + actuator
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 30/50
30
physics vs. sizegravity
friction
Brownian motion
thermal noise
decoherence
faster
smaller
harder to build
MEMS
molecular
quantum
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 31/50
31
multi-agent control� matches control to physics
² different agents for each scale
� matches control to available info² rapid response to local info
² manager agents: overall coordination� without need for details
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 32/50
32
motivation: biologysocial insects, multicellular organisms, ecology
reliable behavior from unreliable parts
cf. incentive issuesnoncooperative agents
economics, common law, «
examplestermite moundsembryo growth
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 33/50
33
motivation: teams� robot soccer
� insect-like robot teams² e.g., foraging
� MSR robots have² tighter physical constraints
² direct access to neighbor locations� e.g., no need for vision to find neighbors
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 34/50
34
topics� modular robots
� multi-agent control
� results² computational ecology
² Proteo
² Telecube
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 35/50
35
computational ecology� dynamical behavior of simple agents
² asynchronous, local decisions
² delays, imperfect information
² ´mean-fieldµ statistical theory² see B. Huberman, The Ecology of Computation, 1988
� apply to actual robot behaviors² see K. Lerman et al. in Artificial Life, 2001
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 36/50
36
techniques� finite-state machine for each module
² simple script, some randomness
� local communication² create gradients through structure
� ´scentsµ
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 37/50
37
topics� modular robots
� multi-agent control
� results² computational ecology
² Proteo
² Telecube
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 38/50
38
example: grow
ing a chain� modes:
² SLEEP, SEARCH(red), SEED( yellow), FINAL(white)
² initially: all in SLEEP, randomly pick one SEED
� seed:² picks growth direction
² emits scent� attracts modules
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 39/50
39
grow
ing a chain
SLEEP
SEARCH SEED
FINAL
descend gradient
+ propagate scent emit scent=0
if neighbor is seed
if neighbor
became seed
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 40/50
40
scent
� set S=min(neighbors)+1
� move around neighbor until lower
value found� if seed found: become new seed
SEEDS = 0
S-1S
S+1
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 41/50
41
structures� recursive branching
² multilevel arms
� grow around object² using contact sensors
See H. Bojinov et al., Multiagent Control of
Self-reconfigurable Robots, 2000
www.arxiv.org/abs/cs.RO/0006030
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 42/50
42
topics� modular robots
� multi-agent control
� results² computational ecology
² Proteo
² Telecube
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 43/50
43
locomotion� make snake shape� move toward goal
² barrier� follow wall� find gap
� higher-level control: general direction² building on low-level agent behavior
see: Kubica et al, Proc. ICRA 2001
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 44/50
44
object manipulation� exert forces to move
object
² based on contact withobject
² ´scentµ recruits
other modules� modules on surfaceform rigid shell
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 45/50
45
summary� simple agents perform basic tasks
² reconfiguration
² locomotion
² manipulate objects
� apply to different hardware types
² Proteo: surface motions² TeleCube: internal motions
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 46/50
46
future directions� quantify capabilities
� design more complex behaviors
� implement on hardware
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 47/50
47
quantify capabilities� examples of agent-based control
² are only specific instances
� quantify² how robust?
² how accurate?
² what cost?� e.g., power use
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 48/50
48
agent design� combine with higher-level agents
² e.g., switch among low-level behaviors
� automate agent design² e.g., genetic algorithms (FXPAL)
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 49/50
49
test on hardw
are� various existing robots
² few, fairly large modules
� large number of tiny modules² don·t yet exist
� wait for hardware vs. simulate?² understand likely hardware capabilities
² e.g., MEMS, «
8/6/2019 Robotics 5
http://slidepdf.com/reader/full/robotics-5 50/50
50
conclusions� agent-based control for MSR robots
² gives robust low-level behaviors
² simplifies higher-level task control
� biological system models² suggest module rules
² useful even if not biologically accurate
www.hpl.hp.com/shl/people/tad