1. CS326 1 An Intelligent User Interface with Motion Planning for 3D Navigation Tsai-Yen Li and Hung-Kai Ting Computer Science Department, National Chengchi University Taipei, Taiwan Based on slides by Shih-Chung Jessy Kang Modified and presented by Vishal Srivastava

2. CS326 2 3D Navigation Interactive 3D graphics is becoming popular on desktop personal computers. Paper deals specifically with low frame- rate, architectural walkthrough applications. VRML browser WALK mode

3. CS326 3 Difficulties in Current Technology Difficulties in 3D Navigation:Difficulties in 3D Navigation: –Using 2D mouse to navigate in 3D environments is difficult for novice users. –Precise navigation control is difficult to achieve with low frame rates –Collision Detection Users often get “stuck” in certain locations and must move backward before continuing

4. CS326 4 Traditional 3D User Interface Various control devices exist, such as Head- mounted displays, Joysticks, and “Eye-balls- in-hand”, but we assume the user has only a mouse.Various control devices exist, such as Head- mounted displays, Joysticks, and “Eye-balls- in-hand”, but we assume the user has only a mouse. Current paradigm for navigating 3D environments with a 2D mouse interfaceCurrent paradigm for navigating 3D environments with a 2D mouse interface 1.User drags a velocity vector v 2.Decompose v into v x, v y 3.V x -> rotational velocity 4.V y -> linear velocity 5.Move view-point (camera) V VxVx VxVx VyVy VyVy

5. CS326 5 Traditional 3D User Interface Problem with the standard Mouse Control paradigmProblem with the standard Mouse Control paradigm 1.If there is potential collision, then stop 2.User may be stuck in certain location, and need to move backward first to escape

6. CS326 6 Assumptions First person view Frame-rate is low (about 10 fps) Camera stays on a horizontal plane –Obstacles can be treated as 2D polygons Environment does not change during run-time –We can pre-process the environment

7. CS326 7 Approach 1.Predict the user’s intended movement 2.Find an appropriate substitute if necessary 3.Plan a path to the destination Integrate a randomized roadmap planner in the 3D navigation

8. CS326 8 Intelligent 3D User Interface Predict User’s Intended destinationPredict User’s Intended destination A.No modification: A1, A2 – legal B.Direct modification: B1, B2 – illegal, but legal along path C.Indirect modification: C – illegal, no legal position along path

9. CS326 9 Intelligent 3D User Interface (cont.) Computing Smooth Maneuver PathComputing Smooth Maneuver Path A.Trivial path: straight line path, A1 B.Non-Trivial path: no straight line, A2 Path Planner is called C.No path: No actions

10. CS326 10 Randomized Roadmap planner A “many-shot” planner that is used in real-time and assumes few changes to the environment “Learning phase” (Preprocessed) 1.Pre-compute C-space with linear-time algorithm. 2.Store in 3D bitmap 128x128x128 3.Divide C-space into 8x8x8=512 regions. 4.Randomly sample up to 4 free configurations in each region. 5.Create a connective graph –Connect all pairs of nodes in the same or neighboring regions with collision-free straight-line paths

11. CS326 11 Randomized Roadmap planner “Query phase” (Run time) 1.Predict intended destination of the user. 2.Connect initial and final positions to nodes in the graph 3.Use a modified A* to create a path connecting those nodes by concatenating path segments. 4.Smooth out the path. 5.Move viewpoint along the path User can intercept the execution of a path

12. CS326 12 Experiment Setting Modify VRML browserModify VRML browser based on Java3D SDK library mouse events 1.The routine for processing mouse events viewpoint configuration 2.The routine for updating next viewpoint configuration Regular PCRegular PC with a Celeron 300A processor WALK modeWALK mode in VRML browsers 10 users –6 CS major students –2 are familiar with VRML browser –2 do not use computers regularly.

13. CS326 13 6 checkpoints in maze-like environment A bouncing ball helps the user identify checkpoints A 2D layout is provided for the testers Experiment Setting 2D-layout map of the maze VRML browser Top View of Maze Environment

14. CS326 14 Experiment Result Planner-generated non-trivial paths

15. CS326 15 Experiment Result Comparison of navigation efficiency with and without path planning Performance speed up of about 73% Save 1/3 navigation steps Substantially more computation per frame

16. CS326 16 Experiment Analysis 1. Number of steps generated by the planner is high (n3 is 1/3 of n2) 2. Extremely high percentage of planned paths are cancelled (n5 is close to n4)

17. CS326 17 Conclusion A path planner with a randomized roadmap approach is used to assist a user in navigating through 3D environments. Concerns –How expensive would the pre-computation be in more complex environments? –Is this useful for applications with high frame-rates? –How useful is the data about total execution time? It would depend on how fast the camera goes along the planned path