Assembly Planning for Mobile Manipulators
This page contains videos and supplemental materials for a paper submitted to the workshop Bridging the Gap Between Task and Motion Planning at ICAPS 2009.
A Hybrid Assembly Task Planning System:
Where Motion Planning Helps Symbolic Planning
Find Good Solutions For Real-World Applications
Overview
Assembly planning for mobile manipulators in complex environments is a difficult task. It requires a planner to consider the inherent step-by-step structure of the problem, as well as the robots' motions required to execute individual assembly steps. Our paper presents an assembly planning framework that enables the automatic generation of assembly plans given a desired structure to be assembled. As execution-time failures occur, the system can seamlessly repair and re-plan the assembly task sequence as necessary.
Below are a number of videos of the (real and simulated) robots in action. We start with an assembly task we have demonstrated using real robots and a scripted assembly sequence. The same assembly can be planned automatically using our planner. Increasing the size of the structure, we show more simulation examples and a mock video of real robots assembling the foru-square lattice structure shown in the picture.
Real Robots -- Single Square -- Scripted Assembly
A video of a team of robots assembling a simple square structure of four nodes and four beams. This video was shot in 2006, and it shows the robots as they execute a hand-written script of tasks. It shows a simple example scenario of the kinds of tasks our planner can work for.
(direct link to movie if embedded version does not play)
As the video shows, our prior work focused primarily on the task execution part of the problem. The robots progress through several behaviors in the process of assembling the structure, and at times, human input is required to recover from failure conditions.
Simulation -- Single Square -- Automatically Planned Assembly
The same assembly scenario as above, but with simulated robots and an autonomously planned task sequence. As the robot encounters execution-time failures, colored boxes mark plan repair and re-planning instances.
(direct link to movie if embedded version does not play)
A yellow marker indicates a (randomly triggered) exception where visual markers necessary for an alignment or installation task are not in view of the camera. A magenta marker indicates an exception during a manipulation task. After three contingency attempts where the robot simply tries again with te same parameters (this solves many of the problems), a plan repair is triggered where the robot backs up and attempts to complete the current task again with new parameters. In this video, the task successfully completes without requiring higher level re- planning. However, if several plan repair iterations had not been able to find a solution to allow the robot to continue, the planner would have been called for help to find an alternate assembly sequence to achieve the goal from the state where the failure occurred.
Simulation -- Two-Square Lattice -- Automatically Planned Assembly
A larger structure of 13 components being assembled in simulation. This two-square lattice is the smallest structure containing an internal component that can be assembled using the components we have available. The planner needs to ensure that the internal component gets assembled while it is still accessible, i.e., before all the surrounding components are put in place.
(direct link to movie if embedded version does not play)
Real Robots -- Four-Square Lattice -- Scripted Mock Assembly
A mock video of the largest structure we can work with given the hardware we have. Our paper describes methods that can be used to automatically generate assembly plans for any sub-set of this four-square lattice.