PILOT (Programming and Interpreted Language Of actions for Telerobotics) is a high level language dedicated to the remote control of systems. Our team has built a complete control system for PILOT, which comprises six main modules: a human−machine interface, an interpreter, a rules generator, an evaluator, an execution module and a communication server. In this paper, we focus on the interpreter which is one of the most important parts of the system. For the initial release of the control system software, the main goal was to have a working environment in order to validate the concepts of the language PILOT. The various modules of the control system have been modeled by finite state automata and the code has been written manually. Although the experimentation carried out with the first release of the control system highlighted the benefits of PILOT for the control of mobile robots such as the robot MARC'H built by our team, some malfunctions were observed in the software and particularly in the interpreter. This paper aims at presenting the work performed, essentially based on testing, in order to detect and to correct software errors into the interpreter, at both conceptual and implementation levels. It ends by our ongoing work related to the use of Petri nets for modeling and testing the interpreter algorithms, the final goal being the generation of safe programs from "validated" models.