In over time.3-Testers knowledge of the context: To

In this section we look at to the new method in the market which is defined forAutonomous Robots. It describes the process of conducting tests while maximizingsafety for operator, robot and other equipments.Robots’ tests should be repeatable and reusable and conducted in a way to maximize safety.Risks to harm human operators, or to damage the robot or other equipments should be minimal. In this test process, three test dimensions are identified:1-The robot activity (Sensing Vs. Acting): A robot can either sense its environment,or it can sense the environment while acting. since acting relies on feedback obtainedthrough sensors, tests involving actuators requires fully functional sensors.2-Context Area (static Vs. dynamic): The context could be static or dynamic. Astatic context does not change while the robot performs its mission. A dynamiccontext is one that evolves over time.3-Testers knowledge of the context: To test a robot, tester should know the environment (e.g. dimensions of the arena as well as locations and sizes of obstacles) withsome levels of accuracy. Based on this knowledge, testers can predict values of senseddata that reflect a correct behavior of the robot. Still, to test a robot more thoroughly, testers need to experiment with different contexts, including ones for whichthey have little knowledge or imprecise one.To help understanding the behavior of the robot, tests should be first done inan context totally controlled by developers. It’s easier to detect faulty behaviorsand trigger security halts. Once a robot passes those tests, testers can proceed withexperiments in less controlled environments.This is the case of tests which check sensing capabilities in an unknown contextthat can be either static or dynamic. Indeed, since the context is not controlled bydevelopers, it makes it difficult for them to define meaningful tests, since they cannot predict valid values r sensed data.Figure 2.1 shows the resulting tests order organized in five safety levels. Upperlevels are prerequisites for lower ones. This means that a test should not be performedon a robot until all tests in the levels below pass.111.3. STATE OF ART CHAPTER 1. INTRODUCTIONFigure 1.8: Three safety test levelsWe now present each test level:1- Testing robot sensing capabilities in a static known environment. This test suiteensures that data produced by sensors matches the expected accuracy level.2- Testing robot sensing capabilities in a dynamic known environment. Such anenvironment is controlled by developers, even if it is dynamic. For example a roomwith some moving entities following a known path or with the light changing accordingto some known patterns. Compared to sensing in a static environment, a dynamicone introduces the time factor. In a dynamic environment, tests are more about thefrequency of collected data than their accuracy. For example, tests in this level mayevaluate if a robot can collect data fast enough to detect potential obstacles.3- Testing robot action in a static known environment. Here the goal is to ensurethat the robot actuators work properly, and that the robot performs basic actions.Still these tests can make use of sensors, since the robot is likely to behave based onperception. Besides, testers should ensure that they have some solutions to stop therobot in case of emergency (e.g. an emergency stop button).4- This level contains two complementary kinds of tests related to robot action inboth static unknown and dynamic known environments.. Testing robot action in static unknown environment. The goal of these tests is toincrease confidence in the tested robot, by making it faces many different situations.12CHAPTER 1. INTRODUCTION 1.3. STATE OF ARTThus, testers have more chance to detect unexpected or unwanted behaviors in situations uncovered by tests in a static known environment. An example of such tests,is putting a vacuum cleaner robot in a room full of obstacles and check that it doesnot get stuck in between.. Testing robot action in a dynamic known environment. At this stage, testers canpredict sensed data. Since the environment is supposed to be known, they can alsopredict when events occur and test if the robot performs the right actions. Tests donot only ensure that the robot does the right action, but they also verify that thebehavior is fast enough. For example, knowing trajectories of all entities of a givenenvironment, one can test obstacle avoidance in a dynamic environment.5- Testing robot action in a dynamic unknown environment. This level representsthe final tests. It aims at confronting tested robots to as many different situations aspossible, especially ones unforeseen by testers. Tests are considered as passed if therobot can cope fast enough with encountered events. An example of a such test ismaking a mobile robot search for a person based on face recognition, in a place manypeople passing by (e.g. a bus station)