Prophetic goal-space planning for human-in-the-loop mobile manipulation

Abstract

This paper introduces the Prophetic Goal-Space Planner (PGP) system developed to accomplish mobile manipulation tasks. Goal-space planning exploits the under-constrained and variable nature of many real-world tasks by defining Cartesian goals in terms of intuitive solution manifolds (e.g., the principal axis of a cylindrical handle, the rim of a valve, the surface of a step) rather than precise points in 6-dimensional space. The PGP system combines 1) a goal-space planner, 2) a kinematic planner and visualizer called the prophet, and 3) user interface tools. The prophet determines appropriate stance locations for the robot to reach its goals and animates the result. The UI tools allow a human operator to either approve or reject the planner's output (the "prophecy"), request new plans, or adjust the goal-space target regions to better match the robot's perceived environment. The PGP system thus enables a level of shared autonomy between the robot and operator that can ensure task completion with only limited, corrective operator input. PGP was employed by Team TRACLabs on the Boston Dynamics Atlas robot for the DARPA Robotics Challenge (DRC) 2015 Finals. PGP enabled the efficient completion of multiple manipulation tasks, including opening and walking through a door, walking up to and turning a valve, and throwing a lever-despite degraded network communication channelsand contributed to TRACLabs' placing 9th out of 23 teams in the competition.

Deploying the NASA Valkyrie Humanoid for IED Response: An Initial Approach and Evaluation Summary

“…Combining generic ATs with automatic detection of possible grasps [27] might also improve operator throughput. Additionally this feasibility study reinforces the importance of identifying candidate stance locations for manipulation tasks [28], [29]. Future autonomous stance generation would preferably reason about collisions, manipulability, maximum exertable forces, and contact constraints.…”

Human-in-the-loop Pose Estimation via Shared Autonomy

“…Our implementation is built upon the Robot Operating System (ROS) [27]. We use affordance template framework (AT) [14,18], which is based on Rviz [14], to perform object registration tasks 2 .…”

Exploring implicit spaces for constrained sampling-based planning

“…For end-effector constraints, one approach is to plan a path for the end-effector, so constraints can be directly evaluated and adhering poses can be sampled. After planning a path within the workspace, a corresponding path in the robot’s configuration space is generated using inverse kinematics (IK) (James et al, 2015; Rakita et al, 2018; Sentis and Khatib, 2005). However, these methods may not be efficient as re-planning is required if a computed path cannot be mapped into the configuration space of the robot.…”