Retraction-Based RRT Planner for Articulated Models

GAMMA Group

University of North Carolina at Chapel Hill

Retraction-Based RRT Planner for Articulated Models, Jia Pan, Liangjun Zhang, Dinesh Manocha

IEEE International Conference on Robotics and Automation (ICRA 2010), 2010, PDF

 

Results:

Benchmark 1: HRR (40DOF) in Bridge Environment

 

bridge0bridge1bridge2

bridge3bridge4bridge5

Planner results:

Video1: a collision-free path generated by our retraction-based articulated model planner.

Video2: a collision-free path generated by our retraction-based articulated model planner (from another perspective).

We obtain a maximum speedup of 4 times on this benchmark.

 

We also test algorithm on human-like robot with 3 different variations.

1)      Articulated model planner:robot is purely considered as an articulated model. The result will be a collision-free path, but robot may ‘fly’ in the air as no task-related constraints are added.

2)      Decomposition planner: robot is planned by an incremental way. The constrained environments reduce the "fly" artifacts.

3)      Decomposition planner with lower body predesigned: some task-related constraints are added, so planner can give more natural motion.

 

Benchmark 2: Human-like robot (41DOF) Object-Picking

picking1picking2picking3

Planner results:

Video1: a collision-free path generated by our retraction-based articulated model planner.

Video2: a collision-free path generated by our retraction-based decomposition planner.

Video3: a collision-free path generated by our retraction-based decomposition planner, with lower body motion predesigned by animator.

We obtain a maximum speedup of 80 times on this benchmark.

 

Benchmark 3: Human-like robot (41DOF) Object-Placing

placing2placing1placing3

Planner results:

Video1a, Video1b, Video1c: a collision-free path generated by our retraction-based articulated model planner, from 3 different perspectives.

Video2a, Video2b, Video2c: a collision-free path generated by our retraction-based decomposition planner, from 3 different perspectives.

Video3a, Video3b, Video3c: a collision-free path generated by our retraction-based decomposition planner, with lower body motion predesigned by animator, from 3 different perspectives.

We obtain a maximum speedup of 4 times on this benchmark and can succeed when the planner without retraction fails.

 

 

Benchmark 4: Human-like robot (41DOF) Car-Bending

car-bending3car-bending2car-bending1car-bending4

Planner results:

Video: a collision-free path generated by our retraction-based decomposition planner, with lower body motion predesigned by animator.

We obtain a maximum speedup of 2 times on this benchmark and can succeed when the planner without retraction fails.

Related Work @ UNC-CH

Human Motion Planning and Synthesis