The project is subdivided five scientific Work Packages accompanied by a Work Package on project management and one that deals with dissemination, communication and exploitation of results.
Work Package 1 defines the Dreams4Cars system architecture, taking the requirements of the automotive application domain into account. It furthermore defines the specifications for the test vehicles and the cloud simulation environments which are necessary to implement the dream-like simulations. Finally, Work Package 1 implements a product quality assurance mechanism to monitor the quality of the improved artificial drivers produced by simulations within Dreams4Cars
Within Work Package 2 the software that implements the driving agent running on-board, the vehicles will be developed. The agent is bio-inspired, implementing a layered (subsumption) architecture priming many possible actions, an action selection mechanism and forward emulators with semantic annotated event logging. The agent must be capable of accommodating re-configured behaviours as they are produced by Work Package 3.
Work Package 3 is the core of Dreams4Cars, where the dream-like mechanism takes place. Among the other things, these simulations develop methods for generating imaginary situations and for the discovery of new action possibilities, with approaches such as optimal control, generalised motor babbling, motivated learning.
In Work Package 4 the cloud environment to enable the dream-like mechanism (the “sleep” state) will be designed and implemented. The simulation environment receives experimental data from vehicles, runs the simulations and redistributes the improved agents. Furthermore, the experimental vehicles from CRF and DFKI as well as the pilot sites will be prepared.
Work Package 5 prepares, organises and monitors the real world driving sessions (the “wake” states). The agent passes through a series of real world sessions of increasing complexity and develops new skills by means of the simulation technology. Final technical assessment will be carried out to verify the added value of the simulation technology compared to traditional human-directed design.
Work Package 6 is geared to dissemination of Dreams4Cars approach and results to scientific robotics and automotive communities. Communication to the general public is also planned. Exploitation of results, contributions to standardisation and liaison with other projects are ingredients of this Work Package.
Work Package 7 provides the managerial backbone for the project. It sets up and manages the processes to ensure high quality of project deliverables and reports. In addition, data management is organised by this work package.
Phase 1: The project starts with the definitions of the architecture and specification. At month 6, the four research Work Packages WP2, WP3, WP4 and WP5 commence. Task Product Quality Assurance also starts at month 6 and will run in parallel with every development activitiy until month 33
Phase 2: At month 18, WP2, WP3 and WP4 merge and phase 2 begins, which is the cyclic training/evolution of the agent and the assessment and optimization of the algorithms. In phase 2, cycles of real world experiences and “dream” states are iterated to track system progress and several evolution lines are simultaneously developed. WP2 and WP3 run in parallel to assess and tune the runtime and simulation subsystem; WP5 monitors the evolution.
Phase 3: In the last 6 month the project closes with data analysis and the finalisation of exploitation plans.
The following figure depicts the Work Package structure, the interrelation of the Work Packages and indicates the three phases.