algorithm - Random Config Generation for RRT -

i writing code rapidly exploring trees robotic arm movement. have 2 doubts

i) distance metric have use find nearest node in graph? if euclidean distance,how calculate because there 2 links in each arm configuration of robot , have no idea how find euclidean distance in case. how find distance between ade , abc if abc nearest config ade in tree?

ii) how generate random config towards goal because random configs never seem reach goal after 5000 iterations.

thanks in advance.

distance metrics 2 revolute-joint arm

rrt pretty robust (pseudo-) metric choose, quality of trees (and consequently paths) influenced if you've got isn't particularly good. performance overall, metric function supposed fast, i'd try simpler things before move onto more complex.

in case of robot arms number of metrics possible. perhaps simplest use euclidean distance between end effector in 2 configurations. you'll have have working if you're testing planning algorithm.

if you've got full dynamics model of system, other metrics based on energy required move arm 1 configuration perform better.

other metrics based on (joint local) angle swept out @ joints, can derived evaluating path inverse kinematics solver may acceptable - haven't tried in practice. may useful technique know if need implement connect-configurations function.

improving convergence

once you've got metric function working correctly, rrt should just work. however, in practice, you'll need oversample near goal configuration encourage algorithm exploit work done in rest of tree building stage. commonly, sampling goal configuration state 5% probability.