200 likes | 371 Views
Planning Biped Navigation Strategies in Complex Environments. Jiaan Zeng. Problem. Plan goal-directed footstep navigation strategies for biped robots through obstacle-filled environments and uneven ground.
E N D
Planning Biped Navigation Strategies in Complex Environments Jiaan Zeng
Problem • Plan goal-directed footstep navigation strategies for biped robots through obstacle-filled environments and uneven ground. • Conventional 2D planning algorithms designed for wheeled robots would be unable to find a solution.
Solution • This paper models the problem as a standard search problem. • The planner uses an A* search to generate a sequence of footstep locations to reach a given goal state.
Model • InputA map representing the terrain to plan over, an initial and goal state. • OutputIf a path is found the planner returns the solution as an ordered list of the footsteps that should be taken to reach the goal.
Model • State Space The state variables x, y, and θ denote the relative position and orientation of the footstep, and the binary variable s ∈ {R,L} denotes which foot is currently the support foot (right or left).
Model • Successor Function(State Transition) • An upper and lower allowable height change HCupper and HClower, and an obstacle clearance value, clearance, representing the largest obstacle that the transition can step over.
Model • Transitions
Model • Evaluation Function (State Evaluation) f = L(Q)+S(Q, T, Qc)+R(Q, Qg)
Model • Evaluation Function: L(Q) • Location MetricsL(Q) =∑ƜiMi (i=1, …, 5) • M1 how slope the plane is • M2 how roughness the plane is • M3 how stability the plane is • M4 the largest bump • M5 how safety the plane is
Model • Location Metrics ― Plane Fitting
Model • Evaluation Function: S(Q, T, Qc) • Evaluation Function: R(Q, Qg)
Algorithms • BFS (Best First) f=g= L(Q)+S(Q, T, Qc) • A* f=g+h=L(Q)+S(Q, T, Qc)+ R(Q, Qg)
Empirical Results • Environment • Simulation • Physical Robot
Empirical Results • Various Types of Terrain
Empirical Results • Distance to Goal
Empirical Results • TransitionsPerformance comparison of BFS (left) and A* (right) for different sets of available footstep transitions.
Empirical Results • TransitionsPerformance comparison of A* (left) and BFS (right) for increasing numbers of stairs along the path from the initial to goal state.
Empirical Results • TransitionsComparison of the output of BFS versus A* on environments with local minima.
Empirical Results • WeightsCarefully choosing the weights for the different metrics is very important for the runtime of the algorithm
Empirical Results • Online Footstep PlanningVision Processing steps. Raw camera images and resulting 3D Depthmap; Mesh model, planar surface identification, and final walking area map.