Robot Ecology

Constraint-Based Design for Long-Duration Autonomy

Magnus Egerstedt

• Éditeur: Princeton University Press
• Année: 2021
• ISBN:
  9780691230078
• Sujet: Tecnologia e ingegneria
• Sous-sujet: Robotica

A revolutionary new framework that draws on insights from ecology for the design and analysis of long-duration robots Robots are increasingly leaving the confines of laboratories, warehouses, and manufacturing facilities, venturing into agriculture and other settings where they must operate in uncertain conditions over long timescales. This multidisciplinary book draws on the principles of ecology to show how robots can take full advantage of the environments they inhabit, including as sources of energy.Magnus Egerstedt introduces a revolutionary new design paradigm—robot ecology—that makes it possible to achieve long-duration autonomy while avoiding catastrophic failures. Central to ecology is the idea that the richness of an organism's behavior is a function of the environmental constraints imposed by its habitat. Moving beyond traditional strategies that focus on optimal policies for making robots achieve targeted tasks, Egerstedt explores how to use survivability constraints to produce both effective and provably safe robot behaviors. He blends discussions of ecological principles with the development of control barrier functions as a formal approach to constraint-based control design, and provides an in-depth look at the design of the SlothBot, a slow and energy-efficient robot used for environmental monitoring and conservation.Visionary in scope, Robot Ecology presents a comprehensive and unified methodology for designing robots that can function over long durations in diverse natural environments.

II Constraint-Based Control

4 Constraints and Barriers

As discussed in the past chapter, environmental constraints, phrased in ecological terms such as predation, territoriality, foraging, mutualism, and so on, give rise to a rich and sometimes surprising set of behaviors among animals [101, 137, 308, 357, 385]. Additionally, when robots are to be deployed over long time-scales, surviving, i.e, not ending up in a situation from which the robots cannot recover, takes precedence over acting in an optimal fashion, as survival is a prerequisite to thriving. In this chapter, we formalize these observations and, in particular, introduce the technical machinery needed to properly manifest constraint-based control design. The key construct towards this end is that of a Control Barrier Function (CBF), and we will build up towards establishing how robot survival can be expressed both compactly and elegantly through the use of CBFs.

Recall the vague yet suggestive statement from Chapter 1 that contained the essence of constraint-based control design for long-duration autonomy,

$$\begin{array}{c}\mathrm{\text{minimize}}{𝒞}_{\mathit{\text{task}}}\hfill \\ \mathrm{\text{subject to}}x\in {𝒢}_{\mathit{\text{survive}}}.\hfill \end{array}$$

Here, x is the state of the robot, and the survival constraint has to hold at each instant of time. But, without explicitly calling out the decision variable over which the minimization takes place, this formulation is hopelessly ill-fated. As such, one needs to add the control input, u, to the formulation, i.e., incorporate the way control decisio...