Efficient Simulation and Navigation

Meeting Minutes (09/27/2011)

Currently: extensive pipeline for dynamic simulation. Dynamic simulators that handle contact use penalty-based methods for articulated robots (potentially slow / unstable)

Multiple steps in controlling using dynamics simulation:

Proposal, DART library in development by Georgia Tech (Karen Liu and Mike Stilman)

Come up with use cases on the wiki as the first step in discussion

use case: come up with library function calls for dynamic motion planning
use case: more broader definition of simulation: (james marble) needs ability for user to specify differential equations that governs the world as plugins, mathml ish more generic extension, for air traffic control simulation
two body friction problem: how to represent in model representation (examine how they're represented in URDF and SDF and suggest improvements).
how to model "state" of the world: persistent vs. delta state
- how to deal with changing model (e.g. mesh deformation)
- how if need to maintain separation of state and model
general feeling that "replanning" is under-represented in ros
efficient message passing of plans and states into simulation
uncertainty: not represented in most modeling representations
look into sensors definitions in sdf, potentially as a starting point for some generic standard
provide easy access for user benchmarking. useful for research, lcp performance comparison, etc. (for this to work, need both model and states standardized)
provide URDF and SDF (Scene Descriptor from Gazebo) as standalone libraries