1. ECE 7340: Building Intelligent Robots QUALITATIVE NAVIGATION FOR MOBILE ROBOTS Tod S. Levitt Daryl T. Lawton Presented by: Aniket Samant

2. INTRODUCTION REQUIREMENTS OF A MOBILE ROBOT Organize its visual memory about local co-ordinate systems of landmarks as the primary means of defining locations Account for the nature of visual events, and represent poor range and angle measurements Maintain memory structures that associate landmark systems along paths of motion that the robot executed when it saw the landmarks Admit inference processes over visual memory that robustly perform navigation and guidance

3. INTRODUCTION GEOGRAPHICAL INFORMATION FOR NAVIGATION Builds a memory of the environment the robot passes through Contains sufficient information to allow the robot to re-trace its paths Constructs or updates a posteriori map of the area the robot passes through Utilizes runtime perceptual inferences and a priori map data for navigation and guidance

4. INTRODUCTION DEFINITIONS PLACE: Region in space, in which a fixed set of landmarks can be observed from anywhere in the region. VIEWFRAMES: Data in terms of relative angles and angular error between landmarks, and coarse estimates of the range of the landmarks LANDMARK PAIR BOUNDARY: Line connecting 2 landmarks, creating a virtual division of the ground surface. ORIENTATION REGIONS: Regions on the ground determined by a set of LPBs PATHS: Sequences of observations of landmarks, viewframes, LPBs and other visual events HEADINGS: World states created by a robotic actions

5. LANDMARKS DEFINITION OF A LANDMARK Distinctive visual events Define a single direction or a set of directions in a 3-D space Visually re-acquirable Robustly acquired and closely related to the navigation task

6. LANDMARKS DETERMINATION OF LANDMARKS WITH SENSORS Knowing the relation between gravity and positions in an image, orient each pixel with respect to the direction of gravity Filtering out the set of n-most interesting structures above and near the global horizon line Defining a local horizon due to sensor’s orientation to the immediate ground plane

7. LANDMARKS ASSUMPTIONS FOR NAVIGATION Has a wide field of view (complete panaromic) Has access to the position and orientation of the immediate ground plane Has a restriction on the camera motion to be translational

8. LANDMARKS

9. VIEWFRAMES GENERATION OF VIEWFRAMES Relative solid angles between distinguished points found using a sensor centered co-ordinate system Distinguished points obtained 1.Statistically – centroid of the landmark projection 2.Structurally – vertex or high curvature boundary point 3.Perceptually – unique visual feature of the landmark

10. VIEWFRAMES Fixes an orientation in azimuth and elevation Takes the direction opposite to the current heading as the zero degree axis Notations L i = Landmarks viewed from left to right Ang ij = Solid Angle between landmarks i and j θ ij = Planar Angle between landmarks i and j e ij = Angular Error between i and j, due to pan/tilt error

11. VIEWFRAMES No need to know co-ordinates of landmarks to deduce approximate angles and distances No need to know the path between the previous and the current location Localization is a geographic region determined by representing valid set of angles between landmarks

12. VIEWFRAMES

13. ORIENTATION REGIONS Orientation of the LPB is [L 1 L 2 ] if Landmark L 1 is on the left Landmark L 2 is on the right Angle from L 1 to L 2 (left to right) is less than 180 degrees Orientation of LPB (L 1 L 2 ) = +1 if θ 12 < π = 0 if θ 12 = π = -1 if θ 12 > π

14. ORIENTATION REGIONS Number of new orientation regions = N – (number of crossings with multiplicities) + 1 Multiplicity = (number of LPBs crossing – 1) Multiplicity = 1 if 2 LPBs cross If all LPB crossings generated by pairs of landmarks in the viewframe have multiplicity 1, number of orientation regions depend on the number of landmarks Number of orientation regions =

15. LPB formed by two landmarks L 1 and L 2 is LPB(L 1,L 2 ) l[L 1 L 2 ] = cross LPB(L 1,L 2 ) to the left of L 1 r[L 1 L 2 ] = cross LPB(L 1,L 2 ) to the right of L 2 b[L 1 L 2 ] = cross LPB(L 1,L 2 ) between L 1 and L 2 a[L 1 L 1 ] = head towards landmark L 1 ORIENTATION REGIONS

16. VISUAL MEMORY FOR NAVIGATION PURPOSES OF LONG TERM MEMORY (LTM) Represent places in the world the robot has passed through so that the LTM is useful for getting back to those places Facilitate visual re-acquisition of landmarks so that the robot can perform recognition processes more efficiently Respond to queries about spatial relational information between places and/or landmarks of interest Formulate and update map knowledge about the world

17. VISUAL MEMORY FOR NAVIGATION

18. LTM ARCHITECTURE Four databases – a priori terrain grids, a priori cultural feature networks (e.g. roads, rivers), viewpaths, and landmarks. Landmark data – time of acquisition, viewframe it occurs in, pointers to map or grid data tying the landmark to absolute coordinate systems, as well as perceptual data. Viewpath maintenance – the process that attempts to match the re-acquisition of landmarks to infer closeness of visual places in memory. Spatial reasoning module – computes relationships between objects in visual memory in response to tasks and queries from the vision system.

19. PATH PLANNING AND EXECUTION HEADING TYPES Type Specifies the co-ordinate system that the direction components are specified in Metric Heading – correspondence of sensor position to a priori map or grid data Viewframe Heading – headings computed between viewframes that share common landmarks Destination Goals Description of places that the heading is intended to point the robot or vision platform toward A set of absolute world coordinates A viewframe localization An orientation region A set of landmarks

20. PATH PLANNING AND EXECUTION Direction Functions Accept runtime data and return true if heading is maintained or false otherwise Compare the desired heading with the sensor reading for metric headings Measure the relative angle between the heading vector and observed landmarks for viewframe headings Termination Criteria Runtime computable conditions that indicate that if heading continues to be maintained, its direction function can no longer return true Conditions Reach destination goal Cross desired LPBs Recognise set of landmarks and relative angles

21. PATH PLANNING AND EXECUTION

22. QUALITATIVE PATH PLANNING Carry a compass as we move through the environment and mark the direction North relative to observed landmarks Propogate paths outward from start and goal regions Store the initial direction for each adjacent region Choose an adjacent region with a relative compass heading most close to the initial step from the start or goal Repeat the process till we get a path

23. QUALNAV SIMULATOR FUNCTIONS Manipulates an a priori grid and cultural feature data Allows the interactive specification of landmarks and vehicle locations Computes visibility of landmarks in 3D terrain data Helps in 2D/3D conversions Supports a number of color graphics Calculates viewframe localizations relative to visible landmarks

24. QUALNAV SIMULATOR Simulator level keeps track of actual landmark locations and performs line-of-sight calculations Planning level records simulated range and actual error for landmark sightings Execution level simulates actual robot motion and vision

25. QUALNAV SIMULATOR