a grasp-based motion planning algorithm for intelligent character animation maciej kalisiak
DESCRIPTION
Animation Techniques many methods: –motion capture –specific gait models –handcrafted controllers –spacetime constraints –etc. cannot solve our problemTRANSCRIPT
A Grasp-based Motion Planning Algorithm for Intelligent
Character Animation
Maciej [email protected]
Introduction
• human character animation• constrained environments• example problem• related research areas
Animation Techniques
• many methods:– motion capture– specific gait models– handcrafted controllers– spacetime constraints– etc.
• cannot solve our problem
Randomized Path Planning (RPP)
• freespace motion planning• piano mover’s problem• example RPP solution
Combined Approach
• borrow ideas from animation and RPP• starting point: RPP• need to add:
– knowledge of human gaits– notion of comfort– moving while in contact with environment
Simplest “Planner”
• character’s state• repeated perturbations,
i.e., Brownian motion• perturbations move COM• inefficient
Potential-guided Planner
• P(q) = COM’s shortest distance to goal
• solve using gradient descent• analytic gradient computation
not feasible• repeatedly sample q’s
neighbourhood and choose perturbations that result in largest drop in P(q)
Local Minima
• gradient descent stops at any minimum• use “random walks” to escape
– Brownian motion of predetermined duration• use “backtracking” if minimum too deep
– revert to a previous point in solution,followed by a random walk
Deep Minimum Example
Smoothing
• solution embodies complete history of search process
• also, very noisy• a trajectory filter post-process is applied
– removes extraneous motion segments– makes remaining motion fluid
Our Extensions
• grasp points• grasp constraint• comfort heuristic system• gait finite state machine• adapted gradient descent, random walk,
smoothing filters
Grasp Points
• represent potential points of contact
• reduces the grasp search space
• “grasp”: attachment of limb to grasp point
• three types
Grasp Constraint
• some number and type of grasps must always be in effect• the number and type of grasps dictated by GFSM• rest of planner must preserve existing grasps
(gradient descents, random walks, smoothing)
The Gait FSM
• provides distinct behaviours• states represent gaits• edges represent transitions • each edge has associated
preconditions and effects• GFSM consulted after every
step of the gradient descent
Heuristic System• each heuristic measures some quality of q• D(q): overall discomfort, a potential field• assuming a comfortable position consists of
using gradient descent through D(q)
Complete System
Results
Future Work
• 3D• grasp surfaces• arbitrary, non-human skeletons• complex grasping• motion speed control• learning
Contributions
• human character animation algorithmfor constrained environments– grasp point discretization of environment– grasp constraint– comfort modeling using heuristics– gait FSM– adapted RPP algorithms to grasp constraint
~FIN~MPEG movies of results available at:
http://www.dgp.toronto.edu/~mac/thesis
Appendix
(extra slides that might prove useful in answering questions)
Character Structure
bitmap and distance map
Alternate gradient descent view
Motion without Heuristics
Smoothing Algorithm
Need for Limb Smoothing
Limb Smoothing Solution
Implemented GFSM
Implemented Heuristics