Learning models for constraint-based motion parameterization from interactive physics-based simulation

For robotic agents to perform manipulation tasks in human environments at a human level or higher, they need to be able to relate the physical effects of their actions to how they are executing them; small variations in execution can have very different consequences. This paper proposes a framework for acquiring and applying action knowledge from naive user demonstrations in an interactive simulation environment under varying conditions. The framework combines a flexible constraint-based motion control approach with games-with-a-purpose-based learning using Random Forest Regression. The acquired action models are able to produce context-sensitive constraint-based motion descriptions to perform the learned action. A pouring experiment is conducted to test the feasibility of the suggested approach and shows the learned system can perform comparable to its human demonstrators.

Authors

• Zhou Fang
• Georg Bartels
• Michael Beetz

Keywords

• Robot sensing systems
• Vegetation
• Predictive models
• Motion control
• Reliability
• Physics
• Learning Models
• Active Learning
• Motor Control
• Simulation Environment
• Actual Knowledge
• Manipulation Tasks
• Flexible Control
• Environmental Variables
• Sequence Changes
• Parametrized
• Minimization Problem
• Cycle Control
• Observable Variables
• Feature Subset
• Learning Module
• Motion Parameters
• Learning Conditions
• Learning Control
• Types Of Constraints
• Slack Variables
• Task Constraints
• Motion Phase
• Random Forest Regression Model
• Motion Constraints
• Event Intervals
• Key Time Points
• Physical Simulation
• Coordinate Frame
• Training Set
• Aspects Of The Task