Failure Recognition and Fault Tolerance of an Autonomous Robot

Abstract

The purpose of this paper is twofold. The rst purpose is to present important issues in designing fault tolerant systems for autonomous robots. The second, is to present the fault tolerance capabilities we implemented on our autonomous robot. Our approach i s c haracterized by a distributed network of concurrently running processes.To tolerate hardware failures, a set of fault tolerance processes are written for each component. These processes are responsible for detecting faults in their respective component, and minimizing the impact of the failure on the robot's performance. By exploiting concurrency and distributedness, the system monitors, detects, and compensates for component failures silmultaneously. The capabilities of this system have been tested by p h ysically disabling and enabling the robot's sensors and actuators. The system quickly recognizes and compensates for both minor and severe sensor and actuator failures. It tolerates a variety of sensor failures such as decalibration, erroneous readings, and permanent failures. It also tolerates various combinations of failures such a s individual failures, concurrent failures, and accumulative failures. We hope this work will inspire further research in fault tolerant a u t o n o m y.

Global behavior via cooperative local control

“…This increases the con dence in the net sensor output. Ferrell (1994) describes how complementary sensor suites can be used to integrate fault tolerance capabilities into distributed controllers.…”

“…Sensori-motor integration on Hannibal is more involved than these earlier systems 12 because Hannibal has considerably more sensors and actuators. Hannibal also has a comparatively large behavior repertoire including insect-like gaits and rough terrain locomotion as well as extensive failure recognition, fault tolerance, and sensor processing capabilities (Ferrell 1993), (Ferrell 1994). The fully compiled system consists of approximately 1500…”

Safety verification of a fault tolerant reconfigurable autonomous goal-based robotic control system

“…Several fault tolerant control architectures for autonomous systems have been developed in which the control effort is layered to deal with faults on different levels, including low levels of hardware control and high levels of supervisory control [5], [6]. Fault diagnosis can be handled by modeling complex systems as stochastic hybrid systems with modes that account for failure states.…”

Constructing controllers for physical multilegged robots using the ENSO neuroevolution approach

“…An important requirement for many robots in the real world is fault tolerance [10]. For example, hardware failures are likely in a robot operating in hostile environments or exploring another planet.…”