www.chatbots.org/nl. â Messengers: buddy .... study: Kismet. ⢠Video: â K is met & R ich. â http://www.ai.mit.edu/projects/sociable/movies/kismet-and-rich.mov ...
Human R obot Interaction A student-lecture by Thomas D ebray Kalle Fischer M ichael Wiedau Yannick S oldati D aan Bloembergen
O verview • • • • • •
C hatbots [Thomas D ebray] R obots & emotions [M ichael Wiedau] Industrial robot interaction [Kalle Fis cher] D omestic robots [Yannick S oldati] B rain computer interfaces [D aan Bloembergen] C onclusion
C hatbots by Thomas D ebray
C hatbots •
P res ence: – C ompanies : helpdesk, website brand agent, advisor, cus tomer service www.chatbots.org/nl
– M es sengers: buddy
http://s ervices.nl.ms n.com/messenger/robots.aspx
•
Future: – E xplosion of chatbots used as brand agent – E motions & visualisation of bot coversations
•
C ommon S tructure: 2 parts (C ase Based R easoning) – Knowledge B ase: encapsulate "intelligence” – interpreter: communicate with user according to instructions in the Knowledge B ase
“s tate machine”: if x is recognized then do y
Loebner prize contest • • •
Annual contes t - www.loebner.net/P rizef/loebner-prize.html D evelop human-like computer Formal ins tantiation Turing Test "C an a M achine Think? “ : natural language proces sing
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Jos eph W eintraub , Thinking S ystems S oftware Jos eph W eintraub, Thinking S ystems S oftware Jos eph W eintraub, Thinking S ystems S oftware Thomas W halen , G overnment of C anada C ommunications R es earch C enter Jos eph W eintraub, Thinking S ystems S oftware Jas on Hutchens C entre for Intelligent Information P rocessing, University of W estern Australia D avid Levy, Intelligent R esearch Ltd. R obby G arner R obby G arner R ichard W allace R ichard W allace Kevin C opple Juergen P irner R ichard W allace R ollo C arpenter R ollo C arpenter R obert M edeks za Fred R oberts and Artificial S olutions
AIM L
by P rofessor R ichard S . W allace (Lehigh University) - ©ALIC E A.I. Foundation
• Artificial Intelligence Markup Language (X M L based) – E asy to pars e by interpreters
• R eleased under GNU GP L • Based on input-res ponse model • S et of templates : – match user's input s tatement – s elect an appropriate res pons e
• C ontext provision:
– S etting – Tes ting of predicates .
• Limitations:
– verbose and difficult to navigate – no direct s upport for s ynonyms , parts of speech
•
www.alicebot.org/TR /2005/WD -aiml/
AIM L
by P rofessor R ichard S . W allace (Lehigh University) - ©ALIC E A.I. Foundation
•
Example: WHAT IS YOUR NAME My name is . WHAT ARE YOU CALLED what is your name
•
Free /O pen S ource AIM L Implementations : R ebeccaAIM L, P rogram D , C hatterBean, P rogram R , P rogram Q , AIM Lbot, P rogram W, C HAT4D , J-Alice, libaiml, P rogram E , P rogram N, P rogram P , P rogram V, P rogram Y/P yAIM L
A.L.I.C .E .
by P rofessor R ichard S . W allace (Lehigh University) - ©ALIC E A.I. Foundation
• • • • • • • • •
Artificial Linguis tic Internet C omputer Entity Loebner P rize in 2000 and 2001 Natural language interface Zipf dis tribution of ques tions Implementation of AIM L (+- 41 000 pre-programmed res ponses ) Based on E LIZA ps ychiatris t program Knowledge bas e = facts , quotes and ideas S upervis ed learning (botmaster manages AIM L content) “Targeting” : automatic detection of patterns in the dialog data => new input patterns that do not already have s pecific replies
Jabberwacky AI by R ollo C arpenter
• •
S tarted up in 1988, released on the web in 1997 Approach: – store everything everyone has ever said – find the most appropriate thing to say by contextual pattern matching techniques – no hard-coded rules => reliance on the principles of feedback ( majority of chatbots are rule-bound and finite)
•
Us e: – not meant for use in busines s circumstances (more control over convers ational flow is required) – learn language and context – form of entertainment – Future: voice operated agent (robots, talking pets)
Emotional Robot Interaction by Michael Wiedau
C ontents • C ase study: Kismet • D esign – – –
General Hardware S oftware
• S ozial Interaction – –
R etrieving E motions S howing E motions
• C onclusion
C ase study: Kismet •
D eveloped – – –
•
at M IT-C S AIL In the 1990's until 2000 by R odney Brooks and C ynthia Breazeal
Features: – – –
S tationary robot Has auditory, visual and expressive systems intended to participate in human social interaction
C ase study: Kismet •
Video: –
Kis met & R ich
–
http://www.ai.mit.edu/projects/sociable/movies /kis met-and-rich.mov
Design General • 6 subsystems • low-level feature extraction system extracts sensor-based features from the world • high-level perceptual system encapsulates these features into percepts that can influence behavior, motivation, and motor processes.
D esign Hardware
D esign S oftware
•
The behaviour design
S ocial interaction R etrieving emotions
S ocial interaction R etrieving emotions
S ocial interaction S howing emotions
S ocial interaction Amplification
S ocial interaction E motions
•
E motion system interacts with High Level P erceptual S ystem and behaviour system
C onclusion • • • • •
C omplex hardware and software system R eacts on human emotions and expressions Humans treat it like a person G ood research project, but finished E xperiences with Kismet influenced many other research projects on robots
R obots in the Industry by Kalle Fischer
• R obots are used by the industries to assemble, weld, paint, pick and place, distribute, etc.. S o for repetitive tasks that do not require human supervision of manpower. • Assets: – can produce the s ame quality for every piece they work on – work fas ter and more precis e than humans – D on't need breaks and do not want more money
R obots in the Industry • D rawbacks: – C an hurt people – C ons ume a lot of energy – Need s pecialized pers onnel
Industrial R obots • D efinition: automatically controlled, reprogrammable, multi-purpose manipulator programmable in three or more axes ( http://www.iso.org/ )
Automated G uided Vehicle (AG V) • D efinition: An AG V is a vehicle that performs transportation tasks in a known environment unsupervised.
Interacting with the robots • S ince there is the big hazard of people getting hurt by robots , the HR I is mostly done from the distance to program the robots. • Industrial robots are mostly programmed for one specific task which is then performed over and over again. R eprogramming is not needed often. • AVGs are programmed to be autonomously and their interaction with humans is res tricted to avoiding to run them over and react to their needs that are communicated using a computer
Interaction Indus trial robots : There are s everal ways to interact or reprogram an industrial robot. M os t commonly, the programming us ing coordinates is us ed. D efine P os itions that the robot s hould manoeuvre its tool to and define action movement and velocity from there on D o this for all subtasks that have to be performed per piece and repeat the sequence over and over programming the robot us ing direct interaction with the robot. P hys ically or electronically lead the robot through the tas k it has to perform and let it reiterate the movement.
Interaction • AVG s: – AVGs mos tly interact with humans by detecting them and avoiding them. – If they are directly interacting with humans that is via a computer that then communicates the task to the AVG.
Interaction with domestic robots by Yannick S oldati
D omestic robots • D omestic robots are designed to assist humans • The 2005-2008 sales projection for all types of domestic robots was 4.5million units, with an estimated value of US $3 billion
HI with autonomous domestic mobile robots
Vacuum cleaner • R ound-shaped – Arms needed
• • • •
Q uantity of dust sensor Auto charging C leaning scheduled Automatically empty its dust
Vacuum cleaner
NEW iR obot R oomba® 610 Profes s ional S eries
• Heavy duty Self-charging Home Base® • 2 Virtual Wall® Lighthouses • 1 Vacuuming Debris Bin • 1 High Capacity Sweeper Bin • 1 Accessory Kit • Rugged Accessory Case • 3 Cleaning Modes • 2-year Manufacturer's Warranty
Lawnmowers • Localization – – – –
GP S S onars C ameras Laser scanners
• A region-filling algorithm is considered in based on neural networks
Lawnmowers • R ain sensor • Wet-grass sensor • Auto-charging battery
Automower from Hus qvarna. Hobbyis ts us e autonomous lawn mowers like the Automower as an inexpens ive platform for building mobile robots .
R obomower from FriendlyR obotics . The R obomower us es V-s haped patterns for mowing . When the R obomower detects an obs tacle, it s tops and turns .
C onclusions • Lawnmowers are still expensive (1190$) – C an be bought for research purpose
• C heapest cleaning robot cost only 130$ from irobot but it is not very efficient for a house • The future is also promising for those who wish to modify a vacuum cleaner to bring tools, serve drinks, or act like a guard dog
B rain-C omputer Interfaces by D aan B loembergen
What are B C I? • B rain-C omputer Interfaces create a bridge between brain and outside world – R eplacement, or prosthes is , for the motor system – C ompensate for los s of control – Two-way: • Brain Actuators • S ensors Brain
Types of B C I • Invasive – Implanted directly in the brain to monitor single neurons or small clus ters of neurons – Low noise, high resolution, but also high risk
• Non-invasive – M eas uring on the skull: E E G – Low ris k, but s ens itive to nois e, low res olution, low bandwidth
Types of B C I • D irect – M eas uring intended actions by looking at the corres ponding neurons – Advantage: no learning required
• Indirect – M apping unrelated but dis tinguis hable intentions onto the true actions – E asier to accomplis h in a non-invasive way
Applications of B C I • • • •
C ommunication R estoration of vision or hearing Limb prosthesis C ontrolling robotic devices – P rovide mobility to paralyzed patients – E .g. “brain-actuated intelligent wheelchair” – Balance between autonomy and control
E xample: Wheelchair • S hared control system (G alán et al, 2008): – Us er provides high level control: directions – Wheelchair handles bas ic tas ks • O bstacle avoidance • Low level control
E xample: Wheelchair • P roblems for AI: – Autonomous “s mart” wheelchair: • Localization • P lanning
– Find balance between autonomy and control
R eferences •
D onoghue, J.P . (2008). B ridging the Brain to the World: A P erspective on Neural Interface S ys tems . Neuron, 60, 511-521.
•
G alán, F., et al. (2008). A brain-actuated wheelchair: As ynchronous and Non-Invas ive Brain-C omputer Interfaces for C ontinuous C ontrol of R obots . C linical Neurophys iology, 119, 2159-2169.
•
Lebedev, M .A., & Nicolelis , M .A.L (2006). Brain-M achine Interfaces : P ast, P res ent and Future. Trends in Neuros ciences , 29(6), 536-546.
•
Vanacker, G ., et al. (2007). C ontext-Based Filtering for As s is ted Brain-Actuated Wheelchair D riving. C omputational Intelligence and Neuros cience, 2007, Article ID 25130.
HR I - C onclusion • C hatbots usage is growing for commercial usage • Interpretation and reaction on emotions makes robots very personal • New D evelopments makes industrial robot interaction saver • D omes tic robots getting much cheaper and more powerful • BC I in combination with an intelligent device could provide mobility to paralyzed patients
Thanks! • Thanks for your attention. • Any questions?