Presentation is loading. Please wait.

Presentation is loading. Please wait.

Behavior-based AI Amir massoud Farahmand

Similar presentations


Presentation on theme: "Behavior-based AI Amir massoud Farahmand"— Presentation transcript:

1 Behavior-based AI Amir massoud Farahmand SoloGen@SoloGen.net Farahmand@ipm.ir

2 Happy birthday to Artificial Intelligence 1941 Konrad Zuse, Germany, general purpose computer 1943 Britain (Turing and others) Collossus, for decoding 1945 ENIAC, US. John von Neumann a consultant 1946 The Logic Theorist on JOHNNIAC--Newell, Shaw and Simon 1956 Dartmouth Conference organised by John McCarthy (inventor of LISP) The term Artificial Intelligence coined at Dartmouth--- intended as a two month, ten man study!

3 HP to AI (2) ‘It is not my aim to surprise or shock you----but the simplest way I can summarize is to say that there are now in the world machines that think, that learn and that create. Moreover, their ability to these things is going to increase rapidly until........…’ (Herb Simon 1957) Unfortunately, Simon was too optimistic!

4 What AI has done for us? Rather good OCR (Optical Character Recognition) and Speech recognition softwares Robots make cars in all advanced countries Reasonable machine translation is available for a large range of foreign web pages Systems land 200 ton jumbo jets unaided every few minutes Search systems like Google are not perfect but very effective information retrieval Computer games and autogenerated cartoons are advancing at an astonishing rate and have huge markets Deep blue beat Kasparov in 1997. The world Go champion is a computer. Medical expert systems can outperform doctors in many areas of diagnosis (but we aren’t allowed to find out easily!)

5 AI: What is it? What is AI? Different definitions –The use of computer programs and programming techniques to cast light on the principles of intelligence in general and human thought in particular (Boden) –The study of intelligence independent of its embodiment in humans, animals or machines (McCarthy) –AI is the study of how to do things which at the moment people do better (Rich & Knight) –AI is the science of making machines do things that would require intelligence if done by men. (Minsky) (fast arithmetic?) Is it definable?! Turing test, Weak and Strong AI and …

6 AI: Basic assumption Symbol System Hypothesis: it is possible to construct a universal symbol system that thinks Strong Symbol System Hypothesis: the only way a system can think is through symbolic processing Happy birthday Symbolic (Traditional – Good old-fashioned) AI

7 Symbolic AI: Methods Knowledge representation (Abstraction) Search Logic and deduction Planning Learning

8 Symbolic AI: was it efficient? Chess [OK!] Block-worlds [OK!] Daily Life Problems –Robots [~OK!] –Commonsense [~OK!] –… [~OK]

9 Symbolic AI and Robotics Functional decomposition Sequential flow Correct perceptions is assumed to be done by vision-researched in a “a-good-and-happy-will-come-day”! Get a logic-based or formal description of percepts Apply search operators or logical inference or planning operators Perception Task execution Planning World Modelling Motor control sensors actuators

10 Challenges of real robotic system Sensors and Effectors Uncertainty Partial Observability of Environment Non-stationarity

11 Requirements of control system of an intelligent autonomous mobile robot Multiple-goal Robust Multiple-sensors Extensible [Learning]

12 Behavior-based approach to AI Behavioral (activity) decomposition [against functional decomposition] Behavior: Sensor->Action (Direct link between perception and action) Situatedness Embodiment Intelligence as Emergence of …

13 Behavioral decomposition build maps explore avoid obstacles locomote manipulate the world sensors

14 Situatedness No world modelling and abstraction No planning No sequence of operations on symbols Direct link between sensors and actions The world is its own best model

15 Embodiment Only an embodied agent is validated as one that can deal with real world. Only through a physical grounding can any internal symbolic system be given meaning

16 Emergence as a route to Intelligence Emergence: interaction of some simple systems which results in something more than sum of those systems Intelligence as emergent outcome of dynamical interaction of behaviors with the world

17 Behavior-based design Robust –not sensitive to failure of particular part of the system –no need for precise perception as there is no modelling there Reactive: Fast response as there is no long route from perception to action No representation

18 A Simple problem Goal: make a mobile robot controller that collects balls from the field and move them to home What we have: –Differentially controlled mobile robot –8 sonar sensors –Vision system that detects balls and home

19 Basic design move toward ball move toward home exploration avoid obstacles

20 A simple shot

21 Different Controlling Mechanism Reactive –Fast, Suitable for structured & a priori known environment –No internal representations Deliberative –Not practical (uncertainty, huge search spaces and …) Hybrid Behavior-based

22 How can we automate it?! How?!

23 Mataric’s Main Trends in Behavior-based Learning Learning Behavior Policy Learning Models of the Environment Learning Models from Behavior History –Michaud Learning Models of Interaction –Goldberg Learning from Human and Other Agents –Nicolescu

24 Learning Behavior Policy They learn (condition,behavior) using RL framework. –No Hierarchy –No MDP Scaling up RL: –Reward Shaping (progress estimator) –Reward Sharing Social Learning –Perceptual Sharing

25 Learning Behavior Policy (2) Mataric, RL in the multi-robot domain, 1997. Mataric, Reward function for accelerated learning, 1994. Mataric, Learning social behaviors, 1997. Mataric, Using communication to reduce locality in distributed multi-agent learning, 1997 Simsarian and Mataric, Learning to cooperate using two six-legged mobile robots, 1995. Others Maes and Brooks, Learning to coordinate behaviors, 1990. Mahadevan and Connel, Scaling RL to robotics by exploiting the SSA, 1991.

26 Learning Models of the Environment They used BBS to learn env. map to show that it is possible for BB structure as a representator mechanism. Map Building Localization Path planning

27 Learning Models from Behavior History Store active behaviors in tree-like structure in which each link shows transition probability. Reduce interference by recognizing common patterns of interference. Michaud and Mataric, Learning from history for behavior- based mobile robots in non-stationary conditions, 1998. Michaud and Mataric, Representation of behavioral history for learning in non-stationary conditions

28 Learning Models of Interaction Augmented Markov Model It is a kind of HMM without any hidden state. Used in order to model behavior transition probability

29 Learning Models of Interaction (2) Why should we use AMM? –No a priori knowledge from env. –Local sensing (need time to estimate env.) –Non-stationary variability How can it help? –Deriving useful statistics from AMM such as mean first pass between two behavior

30 Learning Models of Interaction (3) In a demining task with different values of mine, extract demining time with AMM in order to maximize reward. Multi-agent –Individual performance evalutation (fault detection) –Group affiliation –Group performance

31 Learning Models of Interaction (4) Dani Goldberg and Maja Mataric, Augmented Markov Models (Tech. Report) Goldberg and Mataric, Coordinating mobile robot group behavior using a model of interaction dynamics, 1999. Goldberg and Mataric, Learning multiple models for reward maximization, 2000.

32 Abstract Behaviors Main deficiencies of basic BBS –It has no symbolic and abstract representation –There is no ease of reusability and changing during operation

33 Abstract Behaviors (2)

34 Learning from Human and Other Agents Learning from demonstration Use AB structure

35 Learning from Human and Other Agents (2) Learning is consisted of –Store activated behaviors during demonstration –Set preconditions of NAB

36 Learning from Human and Other Agents (3) Monica Nicolescu and Maja Mataric, Extending behavior-based systems capabilities using an abstract behavior representation, 2000. Nicolescu and Mataric, A hierarchical architecture for behavior-based robots, 2002. Nicolescu and Mataric, Natural methods for robot task learning: Instructive demonstration, generalization, and practice, 2003

37 Learning Behavior learning –How should a single behavior act? Structure learning –How should behaviors arranged in architecture?

38 Overview of learning methods common to Compute Science Supervised learning Reinforcement learning (conditioning, or more precisely operant conditioning) Unsupervised learning

39 Reinforcement Learning Agent sense state of the environment Agent choose an action Agent receives reward from some critic Agent learns to maximize its received rewards through time

40 Structure learning Structure representation –Which architecture and How to make it? Structural credit assignment –What was wrong/right in this structure? Development of structure –When should we add a new behavior –…

41 Structure learning: Purely parallel case build maps explore avoid obstacles locomote manipulate the world sensors

42 Structure learning: Purely parallel case Zero order representation –Version 1! [We have not attacked because it is useless] –Version 2! [OK!] First order representation –Version 1! [Somehow very OK, somehow not so! –Version 2! [Not tested!] –Version 1.5! [OK!]

43 Zero Order Version 2 Test case 1 (4 of 4)

44

45 Zero Order Version 2 Test case 2 (5 of 10)

46 First Order Version 1 Test case 3 (4 ordered behaviors)

47

48 Reward variance as a measure of behaviors’ sufficiency (3 and 4 Beh. Of Test case 3)

49 First Order Version 1 Test case 3 (4 ordered behaviors)

50

51 Readings Brooks, “Elephant don’t play chess,” Robotics and Autonomous Systems 6(1,2), 1990 Brooks, “Intelligence without representation,” Artificial Intelligence Journal 47, 1991 Brooks, “Intelligence without reason,” Proc. 1991 Int. Joint Conf. AI Brooks, “A robust layered control system for a mobile robot,” IEEE J. of Robotics and Auto., 1986 Brooks, “A robot that walks: emergent behavior from a carefully evolved network”, Neural Computation 1, 1989 Brooks, Cambrian Intelligence, 1999 Other Brooks works Mataric’s works

52 Juergen Schmidhuber


Download ppt "Behavior-based AI Amir massoud Farahmand"

Similar presentations


Ads by Google