Testing complex integrated robots (CIRs) requires testing several interacting

control systems. This task is challenging, especially for robots

performing process-intensive tasks such as painting or gluing, since their

dedicated process control systems can be loosely coupled with the robot's

motion control.

Current practices for validating CIRs involve manual test case design

and execution. To reduce testing costs and improve quality assurance, one

trend has been to automate the generation of test cases and execute the

test case automatically as part of a continuous integration process.

This thesis makes two main contributions. First, we present a methodology

for the fully automated testing of CIR control systems. Our approach

is based on a novel constraint-based model for automatically generating

test sequences, where test sequences are both generated and executed as

part of a continuous integration process. We call the methodology CATS,

which is short for Constraint-based Automatic Testing of IPS, where IPS

is an abbreviation for ABB Robotics' integrated paint control system.

Second, we present TC-Sched, a cost-eective method for automatic

test case execution scheduling on multiple machines with constraints on

accessible resources, such as measurement devices or network equipment.

TC-Sched is also based on a constraint-based model and is designed to be

integrated with and executed as part of continuous integration process.

The combinations of CATS and TC-Sched represent an ecient method

for quickly validating the critical software components of CIRs. We rst

use CATS to automatically generate test cases. We then use TC-Sched

to optimally schedule the test cases. We show that, when operating in a

continuous integration process, there is a trade-o between the time spent

solving the constraint model and the time spent executing the result of

the model.

To evaluate our approach, we integrated CATS within ABB Robotics'

continuous integration process. A full lab for the automatic testing of

actual embedded control systems was built. The rst version of the model

was introduced two years ago and has been extended several times. For

TC-Sched, the method was tested on several real industrial test suites, in

addition to a large set of randomly generated test suites. The results are

promising and ABB Robotics has decided to implement TC-Sched in a

full-scale setting at its research facilities at Bryne.

iii

In conclusion, the research presented in this thesis shows that solving

constraint programming models as part of a continuous integration process

can generate realistic test cases and ecient test execution schedules. The

research also shows that the solving process can be performed so that the

trade-o between solving time and execution time is optimal.