120 likes | 352 Views
RAPIDLY EXPLORING RANDOM TREES USED FOR MOBILE ROBOTS PATH PLANNING. Jiří Krejsa, Stanislav V ě chet. Introduction Rapidly explo ring random trees RRT for walking robot RRT for wheeled robot Conclusions. Introduction. Path planning – finding obstacle free path from init to goal node
E N D
RAPIDLY EXPLORING RANDOM TREES USED FOR MOBILE ROBOTS PATH PLANNING Jiří Krejsa, Stanislav Věchet • Introduction • Rapidly exploring random trees • RRT for walking robot • RRT for wheeled robot • Conclusions
Introduction • Path planning – finding obstacle free path from init to goal node • Probabilistic roadmaps • - building roadmap of obstacle free nodes • - interconnect nodes when possible • - connect init and goal • - when path is found it is obstacle free • - probabilistic version: random configurations, local planner • problem: connection problem • Randomized potential field • – object represented as a point • - treated as particle under artificial potential field U • - U constructed to reflect locally the structure of free space • problem: choice of heuristic potential function • Possible solution -> Rapidly Exploring Random Trees - RRT
RRT construction algorithm • RRT.init(xinit) • Repeat • For i=1 to CONNECT_CHECK_INTERVAL • xrand = random state • xclosest = GetClosestNode(xrand) • xnew = GenerateNewNode(xclosest , xrand) • xnew = ApplyRestrictions(xclosest , xnew) • If (xnew is OK) • RRT.AddNewNode(xclosest , xnew) • Else • RRT.Trapped • End if • End for • RRT.TryConnectToGoal • Until GoalReached xinit xclosest xnew xrandom
Obstacle free expansion – uniform covering, no bias Nodes = 20 60 1000 Naive tree RRT tree
Step length influence • Increasing step length speeds up the search up to certain point (Dx=170) • Too high step breaks up the search (Dx>210)
Dx= 30 Dx = 100 Dx = 200 Nodes = 3469 Nodes = 876 Nodes = 3215 Path = 169 Path = 57 Path = 31
RRT for walking robot • - Limited resolution for translational and rotational movement • - Translation: Dx corresponds to robot step • - Rotation: candidate node xnew created in direction closest to xrand • in multiples of rotational steps • Simple test robot used – 4 legs with 2DOF each, HS322 servodrives • Further restrictions right only rotation • right only rotation with no straight movement
ares = 10° ares = 20° ares = 10° avalid - not restricted avalid = (0°,90°) avalid = (10°, 90°)
RRT for wheeled robot • Direct path generation • Node state extension: position, orientation AND velocity vector • Free parameters: steering angle, acceleration • New position/orientation calculated from prior node values • Special cases: constant velocity – steering angle only free parameter • Higher number of states – higher computational demands
Conclusions • RRT – solid, fast reliable technique. • Uniform distribution of nodes over the search space. • Useful for number of constraints. • Simply modifiable to include further constraints. • Future work • Two trees approach Acknowledgement: This work was supported by Czech Ministry of Education under project MSM 0021630518 "Simulation modelling of mechatronics systems".