Paper Note of Classical Planners for Tasks with Continuous Operators

Title: Using Classical Planners for Tasks with Continuous Operators in Robotics

Idea:

  • Provide an algorithm for high-level planning which utilized classic planners and low-level executing:
    • Use Skolem symbols to instead continuous operators for high-level, discrete planning,
    • use sampling to estimate the values of Skolem symbols and RRT-based techniques to implement movements representing each action in the plan.
    • get error message from low level then return to high level planner to refine

Contribution:

  • provides a method for synchronizing between continuous and discrete planning layers
  • utilize classical planner when facts and operator effects over continuous variables are not available a priori

Experiment:

  • simulated using OpenRave
  • Pick and Place Scenario:
    • planner: FastForward (well-established)
    • 30 tests over 3 randomly generated environments
    • 80 to 130 configurations
    • the total average time is far less than the precomp-methods, and needs acceptable re-plan
  • Dining Table Set-Up Scenario
    • planner: FastDownward (cost-sensitive)
    • required 230s, including execution
Show comments from Gitment