1	Syllabus change We are no longer going to cover robotics, vision and natural language as advanced topics in optional lectures. Instead, we will substitute a lecture on natural language processing for the planning lecture. Please see the revised syllabus for more details. Problems? See me. I encourage you to give me feedback.
2	Introduction to Advanced AI Topics Vision Natural Language Processing Robotics
3	Homework #2 We are not yet ready to hand out Homework #2. We will probably have it ready for you by next week. The second homework is on constraint satisfaction problems You can either do it as an individual or as two students in a group. If you’re interested in doing the team assignment, you should find a partner either by talking to people in class or by using the IVLE forum.
4	Advanced Topics Overview Agents have sensors and actuators Sensors: Seeing (visual input) Þ Image Processing and Computer Vision Hearing (audio input) Þ Natural Language Processing Actuators: Moving and manipulating Þ Robotics
5	Computer Vision Perception
6	Definition: versus graphics Graphics Have world model W Generate the sensory stimulus from the model S = f(W) Vision Generate the model from the sensors: W = f^-1(S) To think about: f() doesn’t have a proper inverse. Why?
7	Ambiguity in sensory input Girls playing with dollhouses Or giants playing with people?
8	Definition: versus image processing Image Processing A transformation of data to other data e.g., smoothing Computer Vision Reduction in data to a (more useful) abstraction e.g., digit / face recognition
9	Applications Surveillance – can we detect objects or people as they move around our field of vision? Handwriting recognition – from handwritten addresses to barcodes Content based Image Retrieval – query for images using without any text features. “Show me similar pictures” Automated Driving – speaks for itself
10	Natural Language Processing Communication
11	Definition of NLP Examines communication in human languages. Theoretical and practical aspects. Similar to vision, has production and understanding affects Understanding: speech / text to meaning Generation: meaning to speech / text Both processes have inherent ambiguity
12	Not so great newspaper headlines Squad helps dog bite victim. Helicopter powered by human flies Portable toilet bombed; police have nothing to go on. British left waffles on Falkland Islands. Teacher strikes idle kids.
13	Sample Applications Restaurant Query converts English queries into SQL. MS Dictation converts speech into text Babelfish translates Web pages to different languages Summarizing multiple news articles from the web
14	Robotics Planning in the real world environment
15	Getting around Effectors Sensors on effectors? Is the output noisy? Low-level: need to build higher-level abstractions
16	Problems Localization – where am I? Mobile robots but also robotic arms Mapping – what does my environment look like? Moving – how do I get from here to my goal? What type of plan do I have execute?
17	Applications of robotics Robotic Flight – robotic helicopter, unmanned piloting Path planning for exploration Rock climbing, perhaps difficult even for some of us
18	Summary All three areas deal with search: Vision: search for most likely world w given input sensor s Natural Language Processing: given an input utterance / text i, find most likely meaning m Robotics: Localization: given unknown input configuration / location, determine configuration. Planning: given goal g and state s output plan p to reach g from s
19	Summary All three areas use heuristics : Vision: trihedral structure Natural Language Processing: grammars of language, most frequent meanings Robotics: decomposition of problems into cells, maximizing distance between obstacles Many of these heuristics involve probability, which we will return to at the end of the semester.