Transcript lecture23
CS 416 Artificial Intelligence Lecture 23 Computer Animation No class on Wednesday Unfortunately I will be out of town Final Exam The final exam… OLS 009, December 17th at 7:00 p.m. It will not be cumulative Study sheet, midterm answers, and extended office hours will be forthcoming Halo Tournament Halo 2 Saturday, Dec. 11th 6:30 $3 for ACM members, $5 for others Teams of Four Big Prize Details: http://acm.cs.virginia.edu/halo Fun Stuff Why do I love this stuff? Impact of Video Games Annual video game revenues surpass box office Worldwide revenues $6 Million Man $7 Billion Man $5.6 Billion Man Number of transistors on GPU doubles each 6 mos. • Three times Moore’s Law But It’s Motion That Matters But It’s Motion That Matters Animation Techniques Keyframing Animation Techniques Motion Capture Microsoft Motion Capture Group Motion Analysis Michael Gleicher Animation Techniques Procedural Characters Physically Simulated • 12 Rigid Bodies • 17 Controlled Degrees of Freedom • Body Segment Densities from Biomechanical Data • Mass and Moments of Inertia Calculated from Polygonal Model Torques Applied at Joints Rider attached to the seat Forces applied by rider on handlebars Wheels roll on ground without slipping Forces applied by rider on pedals Locomotion Control Correct for errors in: Speed Hodgins, Wooten, Brogan, and O’Brien SIGGRAPH 1996 Orientation Roll Physical Limitations Action Lags • Limited acceleration • Maximum/Minimum velocities Mobility Constraints • Architecture of Actor How Mobile Are You? Build a Model of Human Mobility • Use motion capture • Use simulation • Use biomechanics Modeling Pathologies What is optimal locomotion strategy in presence of neuromuscular malfunction? • Large search space – Limb trajectories – Joint torques – Time to complete movement Brogan, Sheth (Mechanical Eng), and Granata (Biomed) Applied Modeling and Simulation 2002 Spacetime Constraints Witkin and Kass, 1988 • An optimization method – Shoot for the stars – Deal with physics later Human Path Planning Simulate goal-oriented walking in cluttered room • Parameterize simplified human dynamical model • Learning by example Brogan and Johnson (Psyc undergrad) Graphics Interface 2003 (to be submitted) Five Experimental Conditions Track trajectories while moving from A to B • 40 participants, 7 paths each Build Model • Speed Profiles – acceleration and deceleration fit to cubic (distance to goal) – Min/max speed – slow while turning • Turning radius • Subgoals located at closest points to obstacles • Compromise between current and direct trajectories – Inertial dampening Evaluation • Compare to two A* models with multiple lookahead values • Evaluate – piecewise errors in position / velocity – area between paths Image-based Method Image matching is efficient • Video Textures – Schodl et al., SIGGRAPH 2000 Image matching must be carefully normalized • Cross-correlation between current image and all in database • Assign a matching weight to each image in database Simulation Level of Detail Planning actions of bicyclists • Model must include plan for where bicyclist would like to go • Modified by knowledge of what bicyclist can accomplish Brogan and Hodgins AAMAS 2002 Know Your Limits Heuristically Tuned Previous results – obstacle avoidance Selecting an Action Use Simplified Models to Predict Building Simplified Simulations Explore range of actions in many circumstances Store results in compact representation Results Bicyclist Average Tracking Error Results Sum Position Errors Simulate 120 Sec We discuss good papers like these in the animation course • Neuroanimator (Grzeszczuk et al.) • video • Through the lens camera control (Gleicher and Witkin) • Video Textures (Schodl et al.) • video • Flocks, Herds, and Schools (Reynolds) • Escape panic / Pedesetrian Crowds (Helbing) Automatic Motion Capture Estimation (O’Brien et al.)