Whole-body Navigation and Forceful Manipulation of Deformable Objects for Humanoid Robots Performing Shipboard Maintenance Tasks
Abstract
Objective: We envision autonomous robots that perform shipboard maintenance and service tasks, but to be effective in these scenarios, robots need 1) the ability to navigate shipboard spaces subject to sea state disturbances, and 2) the ability to manipulate many kinds of deformable objects (such as a wire, valve components, and cables). This project will focus on two thrusts to develop these abilities: Thrust 1: For navigating shipboard spaces, compliant control at the low-level will allow robustness to small variations in geometry and small disturbances, however accommodating complex geometry and large disturbances requires a unified approach to whole-body motion planning, navigation, and disturbance mitigation. Thrust 2: For manipulating deformable objects, a key barrier to enabling many practical tasks is forceful manipulation, i.e. where significant forces must be applied to the object to accomplish a task. While it is possible to manually create task-specific controllers to perform such tasks, no broad algorithmic framework for forceful manipulation of deformable objects currently exists due to a lack of characterization of possible error modes and methods to escape from states where the object is ~stuck.~Approach: Thrust 1: We propose to develop a new Disturbance-robust Whole-body Navigation (DaRWiN) planning framework. While we will build on our prior work in whole-body motion planning, creating the DaRWiN framework requires overcoming fundamental research challenges. Specifically, we will 1) Create methods that reason about compliance to geometric disturbances in motion planning by studying how compliant controllers respond to different types of collisions; 2) Investigate generating motion plans that are robust to large disturbance forces by planning tobe near footsteps and hand-holds that can mitigate large disturbances; 3) Validate and verify our methods on the SAFFiR robot, both in simulation and with physical tests. We will culminate tests with a demonstration on the ex-USS Shadwell in a simulated fire-suppression scenario. Thrust 2: The idea at the center of the proposed framework is a method to plan belief-space policies thatanticipate when the object is likely to get stuck, jammed, or caught and provide contingencies. These policies will exploit new types of motion primitives to escape the above local minima. We will test and verify the efficacy of these methods on three Navy-relevant tasks: 1) Replacing the o-ring in a valve; 2) Removing/replacing wire from/on terminals for switchboard cleaning; and 3)Routing a fiber-optic cable through a complex environment.Short Statement of Work: We envision completing this work in a four-year time frame. Thrust 1: Simulation and testing on the physical robot at Virginia Tech on tasks involving door-opening and traversing a narrow hallway will take place in year one. Year two will focus on developing our methods for humanoid navigation under disturbances and validate and iterate our methods onthe physical SAFFiR robot performing similar tasks to those described for simulation. Thrust 2: Year one will focus on development of the planning approach for belief-space policies with a focuson investigating the exploration and sampling strategies used in the planner. The second year of the project will focus on online policy-repair methods that can amend the policy during execution when an incorrect prediction occurs. In year three we will test on all three tasks in simulation and perform a physical o-ring replacement on a valve and iterate our methods. In the final year we willfinish the algorithmic development and test on the remaining two tasks in physical mock-ups.ONR/Navy Relevance: Regular maintenance of equipment is essential to reduce the need to replacecostly components and to maintain military readiness. We expect this research to provide a significant step toward autonomous shipboard robots that can perform many routine maintenance tasks that are time consuming, tedious,
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2017
- Source ID
- N000141712050
Entities
People
- Dmitry Berenson
Organizations
- Board of Regents of the University of Michigan
- Office of Naval Research
- United States Navy