Adaptive Tool Use for Shipboard Maintenance Robots 23-000005833
Abstract
Research Problem and Objectives: We envision autonomous robots that perform shipboardmaintenance tasks. To be effective in these highly-constrained and dynamic scenarios, robots needthe ability to adapt to disturbances and variations in the task-at-hand. For example, a robot taskedwith replacing spark plugs on a marine diesel engine will need to use various wrenches to removebolts around the engine casing, remove the spark plugs, and re-fasten the casing bolts. In general,the problem domain we consider is adaptive tool-use for maintenance tasks#where we can sensethe state of the robot and tool being used, but we do not know the exact location of the object beingmanipulated and its frictional properties. To enable adaptive robotic tool use in these scenarios, wewill pursue the following technical approaches:Technical Approaches: We will create a method that replans dynamically in a Model PredictiveControl (MPC)fashion. Many methods have been developed for MPC but the relevant methodsrely on sampling trajectories. Unfortunately, the probability of sampling a trajectory that meets astringent set of constraints, e.g. for a tool maintaining contact during motion, is often 0. In recentwork, we developed a trajectory optimization method, Constrained Stein-Variational TrajectoryOptimization (CSVTO), that can enforce a variety of constraints (such as pose and force for arobot arm) while maintaining a diverse set of high-quality trajectories. The goal of this projectis to build an adaptation framework around CSVTO to make it applicable to shipboard tool-useand maintenance scenarios. The adaptation framework will consist of three parts: adapting goalstatesfor tool-mating, learning end-conditions for guarded moves, and adapting force constraintspecifications from online data. We will test and verify the efficacy of this adaptation frameworkon three common Navy-relevant maintenance tasks: 1) Loosening and tightening bolts in an engineassembly; 2) Replacing spark plugs in an engine; and 3) Changing engine oil.Anticipated Outcome: The expected outcome of this research is an adaptation framework forMPC which can accomplish realistic tool-use tasks. Given a CAD model of the environment and atask specification (e.g. remove a certain bolt), the robot will be able to mate the specified tool withthe object to be manipulated and perform the task. For tasks involving many steps, e.g. changingthe engine oil, the robot will perform a series of specified tool-based manipulations using the aboveframework.Impact on DoD Capabilities: Regular maintenance of equipment is essential to reduce the need toreplace costly components and to maintain military readiness. We expect this research to providea significant step toward autonomous shipboard robots that can perform many routine maintenancetasks that are time consuming, tedious, and/or costly to perform for Navy personnel. Theproposed algorithms, when deployed on shipboard robots, could also make performance of maintenancetasks more consistent and reliable while allowing Navy personnel to focus on other duties.Another application of the methods proposed in this project is to engine maintenance onboard unmannedships. In such cases, robots can be integrated into the engine bay to perform repair andmaintenance tasks. Such robots can use the proposed algorithms to extend the deployment timeof unmanned vessels by performing regular maintenance and simplerepair tasks. We envisionthat robots endowed with adaptive tool-use capabilities will also be effective at naval bases. Suchrobots could be used to perform maintenance on automobiles at the base or on the base#s HVACsystems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 15, 2023
- Source ID
- N000142412036
Entities
People
- Dmitry Berenson
Organizations
- Board of Regents of the University of Michigan
- Office of Naval Research
- United States Navy