< Back to previous page

Project

Exploiting Natural Dynamics for Efficient and Robust Control of Modern Robots: Identification and Realization of Design Patterns

Motivation: Very often, control objectives such as accuracy and precision of robot motions and forces are not the only requirements imposed on the robot's behaviour. For completing tasks defined in dynamic and unstructured environments/settings, it is necessary that the employed software control method additionally enables high degrees of efficiency and robustness in a robot, given the system's mechanical structure and actuation capabilities. It has been shown that the advanced efficiency and robustness of humans and animals, in performing the locomotion and manipulation tasks, are enabled by a control paradigm that exploits (i.e. takes the advantage of) their natural dynamics. Those insights motivate us to consider the same control paradigm, i.e. embrace natural dynamics of the robotic systems, while addressing the problem of enabling such advanced performance for our robots.

Problem formulation: The existing strategies in the State of the Art are implicit and not sufficiently generalizable when it comes to evaluation and exploitation of the natural dynamics. More concretely, those solutions i) do not provide a clear reasoning on which parts of natural dynamics are exploited and how; ii) are tailored to specific mechanisms and task formulations; and iii) in some cases, computationally impractical for real-time applications.

Planned scientific contribution: This PhD study aims at developing reusable control methods for exploiting robot's natural dynamics in unifying, explicit, general and real-time applicable ways. More concretely, those reusable solutions will provide a strong step towards advanced and computationally efficient control algorithms and architectures, and their automatized and formal development. In this context, some of the above-used terms have the following meaning: unifying - considering all the mechanical characteristics that are available in the system; explicit - it is clear which parts of natural dynamics are exploited and in what capacity, during the run-time; general - the derived methodology can be applied on different types of manipulators, mobile platforms and legged robots, i.e. it can generalize over a variety of different robotic structures and actuation capabilities; real-time applicable - allows for achieving the real-time computations.

Main task: It is necessary to derive model-based and/or learning-based design patterns for the control computations that i) explicitly and algorithmically evaluate all characteristics/aspects of the system's natural dynamics, or in other words, resolve all effects that natural dynamics impose on the system, in a holistic and recurrent manner; ii) cleverly utilize (take advantage of) their features in control to improve the performance of a robot, in terms of efficiency and robustness; iii) exploit/leverage structures and representations of dynamics and control laws such that the real-time applicability of derived algorithms and overall architectures is improved.

Date:25 Feb 2022 →  31 Aug 2023
Keywords:Robotics, Dynamics, Energy, Control, Design Patterns
Disciplines:Adaptive agents and intelligent robotics, Control systems, robotics and automation not elsewhere classified, Classical mechanics, Embedded and real-time systems
Project type:PhD project