Robust Space-Time Footsteps for Agent-Based Steering By Glen Berseth 1, Mubbasir Kapadia 2, Petros...
-
Upload
crystal-heath -
Category
Documents
-
view
217 -
download
0
Transcript of Robust Space-Time Footsteps for Agent-Based Steering By Glen Berseth 1, Mubbasir Kapadia 2, Petros...
Environment Optimization for Crowd Evacuation
Robust Space-Time Footsteps for Agent-Based SteeringBy Glen Berseth1, Mubbasir Kapadia2, Petros Faloutsos3
University of British Columbia1, Rutgers University2, York University3 Picture
Have 15 minutes for talk? 12 minutes for presentation 3 minutes for question?1Interface between steering and motion synthesisSimple Sliding DiskPositionVelocityNo information about limbs
DiscsTo large, bad crowd packingTwo small, limb collisions
2GoalsBetter interface between steering and motion synthesisSufficiently detailed information for motion synthesisEfficient space-time planningHeterogeneous agentsBetter qualitative performance
3Contributions:Geometric pruning of search space Search uses randomized step directions and timesNew types of footsteps (In-place turning)Benefits:5x performance increaseImproved stability and local minima avoidanceElimination of certain deadlocking configurations4Related WorkFootstep-based steeringRobotics [Lots]Animation [van Basten and Egges 2010, Singh et al. 2011]The work that is related to this research5Example
6
Footstep State Space
{Com Position, COM velocity, current foot position, current foot orientation, current foot}7Motion PlannerNear-optimal finite horizon best first searchCost function:Maintaining certain speedChanging momentumHeuristic estimate:Estimates the number of steps to the goal and then computes the associated energy cost
Regions are important, dont use goal location8Successor State GenerationAd-hoc, discreteRandomized, continuous
Possible successor states for right foot
Ad-hoc methods are susceptible to complex configurations9Dynamic Collision Model
Dynamic collision model is complex10Geometric Validation
Initial placementEnd of finite horizon plan11Examples
Analysis metrics:Quality:Complete:Reached target locationSolved: Reached target location without collisionEfficiency:SimTime: Time to finish simulationAnalysis
99.7% completion[Singh et al. 11]Describe metrics14Questions?Robust Space-Time Footsteps for Agent-Based SteeringBy Glen Berseth1, Mubbasir Kapadia2, Petros Faloutsos3
University of British Columbia1, Rutgers University2, York University3
Footstep Model
Sliding Disk IssuesHas fixed collision boundariesProvides no contact pointsFew locomotionconstraints (turning radius, sharp changesin direction)No space time planning
17