Search

Embedded software is shaping and influencing our world and it is unimaginable to realise day to day life without it. Since the introduction of the first Electronic Control Unit (ECU) in the 1970s, the automotive industry has seen a substantial increase of embedded software in vehicles. The use of embedded software in the automotive industry has led to a significant increase in the number and complexity of different vehicle systems, features and functions. This level of complexity drives premium vehicles with no fewer than 70 ECUs interconnected by more than five on-board network systems such as Controller Area Network (CAN), Local Interconnect Network (LIN), Media Oriented Systems Transport (MOST), FlexRay and Ethernet. In a typical automotive development process, the main challenge for the engineers is to uncover as many failure modes and/or software defects as possible during the early stages of the vehicle programme. During the early phases of the development, failure modes and/or software defects are difficult to uncover but easy and inexpensive to fix. During the latter phases of the development, failure modes and/or software defects are easy to uncover since the final product has been built. At this stage, failure modes and/or software defects are hard and expensive to fix as changes required in the embedded software. The aim of this research was to develop and deploy innovative solutions in order to shift failure modes and/or software defects detection early in automotive product development. The initial research work was conducted through an analysis of failure modes and/or software defects found during a typical Jaguar Land Rover (JLR) vehicle programme development. This preliminary work also then focused on supplier base capability for automotive embedded software development. The research findings from the internal and external analysis, together with the literature review on best practice have driven the development of four solutions. A process called Model-based Product Engineering (MBPE) was created and deployed within JLR. The MBPE process brings together model-based development and other development processes in a standardised form. A new generic Design Verification Interface (DVI) for test exchange and traceability across all MBPE process levels was developed. The generic DVI eliminates or reduces redundant efforts of re-writing test cases and test scripts for automated testing. A semi-formal Standardised Design Verification Method (SDVM) was developed for defining test cases for all vehicle systems in a common template. The SDVM presents test cases as machine readable data and allows auto-generation of test scripts suitable for automated testing. An end-to-end solution called Platform Independent Test System was developed in order to integrate the MBPE, DVI and SDVM solutions. The proposed PITS supports all levels of system abstraction from the test case definition phase to the execution of automated scripts in both offline and real-time test environments. Evaluation results have demonstrated a significant shift in the detection of failure modes and/or software defects towards the early phases of the product development. An early detection of more than 50% of failure modes and/or software defects was achieved. This is a substantial change from the previous state where embedded software validation was conducted only after supplier software release. Furthermore, results have shown a 40% reduction in engineering effort for test scripts creation and a five to tenfold reduction in engineering time for automated testing.

This thesis is concerned with the development, design and implementation of a novel hybrid multi-agent orientated control architecture for navigation of multiple autonomous mobile robots operating in an unknown and unstructured environment populated by static and/or dynamic obstacles. The proposed hybrid control architecture is modular and draws its design from competitive tasks architecture, production rules architecture, connectionist architecture, dynamic system architecture, multi-agent architecture and subsumption architecture. The reasoning of the control architecture is both deliberative and reactive. The proposed reactive behaviours are modelled using fuzzy logic, neural networks and hybrid behavioural encoding incorporating stateflow-fuzzy logic and stateflow-neural networks. The deliberative system is comprised of finite state machines. The processing is achieved in a centralised and/or decentralised manner using the proposed controller-agent concept from the field of multi-agent systems. The framework of the control architecture is suitable for adaptation in single and multiple robot navigation. The control architecture has been implemented in MATLAB/Simulink using the full non-linear model of the MIABOT V2 mobile robots. It is evaluated incrementally in order to verify its overall control performance and the performance of each subsystem. Results show that the control architecture's modularity, distribution, reactivity and behaviourbased structure provided the overall control system with robustness in all cases of navigation tasks utilising either single or multiple mobile robots. Furthermore the results obtained show the effectiveness of the control architecture in navigation tasks involving up to five mobile robots operating in unknown static and dynamic environments. The results demonstrate that the control strategy chosen for navigation of multiple mobile robots is efficient and also established the robustness of the control system architecture against the desired requirements, such as supervision, decision-making and co-ordination of internal control structures (subsystems). The autonomous mobile robots were exposed to a complex and highly dynamic environment and successfully achieved every control objective. Their trajectories were smooth despite the interaction between several behaviours and the presence of unexpected static and dynamic obstacles. The main contributions of this thesis are: development of a novel hybrid multi-agent based control architecture called CARDS; novel approach for identification of direction of moving obstacles (other robots) using finite state machines; novel approach for behavioural encoding using hybrid solutions such as stateflow-fuzzy and stateflow-neural for autonomous robot navigation; proposed a design methodology for developing integrated solutions for autonomous mobile robotic systems and classification of the main design methodology (properties) of control systems architectures for autonomous mobile robots. Less significance contributions are: literature survey on approaches/methods related to the development of intelligent control architectures for navigation of multiple autonomous mobile robots; modelling of MIABOT V2 mobile robots; comparison between PI, fuzzy and neural controllers and algorithmic methodology for discovery of fuzzy/neural local models from observation data; identification of the relationship of the most important requirements/properties of control architecture versus the main control architecture specifications using the Quality Function Deployment tool; modular approach for modelling and evaluation of three types of sensor and sensor sensitivity.