1. Planning and control of robotic manipulation actions for extreme environments
- Author
-
Pardi, Tommaso
- Subjects
AI Indexes (General) ,QA75 Electronic computers. Computer science ,QA76 Computer software ,TK Electrical engineering. Electronics Nuclear engineering - Abstract
A large societal and economic need arises for advanced robotic capabilities, where we need to perform complex human-like tasks such as tool-use, in environments that are hazardous for human workers. This thesis addresses a collection of problems, which arise when robotic manipulators must perform complex tasks in cluttered and constrained environments. The work is illustrated by example scenarios of robotic tool use, grasping and manipulating, motivated by the challenges of dismantling operations in the extreme environments of nuclear decommissioning Contrary to popular assumptions, legacy nuclear facilities (which can date back three-quarters of a century in the UK) can be highly unstructured and uncertain environments, with insufficient a-priori information available for e.g. conventional pre-programming of robot tasks. Meanwhile, situational awareness and direct teleoperation can be extremely difficult for human operators working in a safe zone that is physically remote from the robot. This engenders a need for significant autonomous capabilities. Robots must use vision and sensory systems to perceive their environment, plan and execute complex actions on complex objects in cluttered and constrained environments. Significant radiation, of different types and intensities, provides further challenges in terms of sensor noise. Perception uncertainty can also result from e.g. vision systems observing shiny featureless metal structures. Robotic actions therefore need to be: i) planned in ways that are robust to uncertainties; and ii) controlled in ways which enable the robust reaction to disturbances. In particular, we investigate motion planning and control in tasks where the robot must: maintain contact while moving over arbitrarily shaped surfaces with end-effector tools; exert forces and withstand perturbations during forceful contact actions; while also avoiding collisions with obstacles; avoiding singularity configurations; and increasing robustness by maximising manipulability during task execution. Furthermore, we consider the issues of robust planning and control with respect to uncertain information, derived from noisy sensors in challenging environments. We explore the Riemannian geometry and robot's manipulability to yield path planners that produce paths for both fixed-based and floating-based robots, whose tools always stay in contact with the object's surface. Our planners overcome disturbances in the perception and account for robot/environment interactions that may demand unexpected forces. The task execution is entrusted to a hybrid force/motion controller whose motion space behaves with compliance to accommodate unexpected stiffness changes throughout the contact. We examine the problem of grasping a tool for performing a task. Firstly, we introduce a method for selecting the grasp candidate onto an object yielding collision-free motion for the robot in the post-grasp movements. Furthermore, we study the case of a dual-arm robot performing full-force tasks on an object and slippage on the grasping is allowed. We account for the slippage throughout the task execution using a novel controller based on the sliding mode controllers.
- Published
- 2022