Download presentation
Presentation is loading. Please wait.
1
Multi-Rep Multi-Heuristic A*
Bipedal Representation Anchor Heuristic : FeetToGoal : Dijkstra heuristic to the goal for 25cm on the ground. This is used for generating the lower dimensional planning footprints. Inadmissible Heuristics : RadialStair : This weights the FeetToGoalHeuristic such that the search prefers expanding the states in the middle of the stairs than on the edge of the steps. h = 1 / (distance from centre of step)2 x Dijkstra of Feet to goal. Ladder Representation Anchor Heuristic : Same as the anchor for bipedal representation.
2
Multi-Rep Multi-Heuristic A*
Full-D Humanoid Representation Anchor Heuristic : RootToGoal : Dijkstra heuristic to the goal for 25cm around the root of the robot. This is used for generating the higher dimensional tracking. Inadmissible Heuristics : COMToTargetPolygon: This heuristic is the sum of the following components h = Euclidean of COM of current state to Centroid of Target Polygon + Number of states left on the Adaptive low-D path HeadingDifference : The angular difference between the feet of the robot and the root. h = angular difference between root and feet.
3
Multi-Rep Multi-Heuristic A*
LiftFeetToTarget : This heuristic is intended to help the humanoid lift its feet from the current state on top of the target footprint h = Euclidean of Current foot position to a lifted Target footprint + Number of states left on the Adaptive low-D path Fitted-Curve for Feet Guidance : Potentially we can fit a parametrized curve between previous footprint and target print such a parabola or a circle which would guide the humanoid to place the feet from one footprint to the other along this curve. h = F(Previous FP, Target FP and Step height) + Number of states left on the Adaptive low-D path Note : Currently it fits a semi-circle between the footprints. Several different curves can also be incorporated as a guidance for the foot in the future, and be used as different heuristics.
4
Multi-Rep Multi-Heuristic A*
Bipedal-HandRail Representation Lower dimensional planning is handled by appending closest handrail contact to each expanded feet as the arm-limb target position. Higher-dimensional planning tracks targets for both arms and feet limbs. Full-D Tracking Heuristic : Euclidean distance between arm footprints + COM distance between current state and Centroid of target polygon. Possible Future Work : Incorporate proper handrailing contact information. Learn set-of-actions / macro-action from database of plans to use as heuristics for arm handrailing. This is also applicable for planning for walking. Derive heuristics for different homotopy classes for handrailing, so that the search can choose the best homotopy class for holding handrails.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.