Navigation Among Movable Obstacles with learned dynamic constraints

In this paper we present the first planner for the problem of Navigation Among Movable Obstacles (NAMO) on a real robot that can handle environments with under-specified object dynamics. This result makes use of recent progress from two threads of the Reinforcement Learning literature. The first is a hierarchical Markov-Decision Process formulation of the NAMO problem designed to handle dynamics uncertainty. The second is a physics-based Reinforcement Learning framework which offers a way to ground this uncertainty in a compact model space that can be efficiently updated from data received by the robot online. Our results demonstrate the ability of a robot to adapt to unexpected object behavior in a real office scenario.

Authors

• Jonathan Scholz
• Nehchal Jindal
• Martin Levihn
• Charles Isbell
• Henrik I. Christensen

Keywords

• Robots
• Aerospace electronics
• Grippers
• Navigation
• Learning (artificial intelligence)
• Dynamics
• Planning
• Movable Obstacles
• Markov Decision Process
• Real Robot
• Angular Velocity
• Friction Coefficient
• Use Of Control
• Large Coefficient
• Target Object
• Manipulation Tasks
• Large Objects
• Linear Velocity
• Dynamic Objects
• Humanoid Robot
• Current Beliefs
• Kinematic Constraints
• Object In Frame
• Steering Control
• Object Trajectory
• Body Velocity
• Monte Carlo Tree Search
• Object Velocity
• Robot Base
• Mobile Manipulator
• Physics Engine
• Tree Search
• Global Status
• Object Motion