iCO2: multi-user eco-driving training environment based on distributed constraint optimization

Multi-agent systems have already been successfully applied to a variety of traffic control problems and demonstrated the potential to lower travel times and environmental impact. Sharing this goal, we have developed iCO2, an online tool for training eco-friendly driving in a multi-user three-dimensional environment. iCO2 supports eco-driving practice by instructing computer-controlled agents, such as traffic lights and other vehicles, to create traffic situations that make eco-driving more difficult. Hence the agents take the role of “opponents” that try to achieve the optimal challenge level for the skill level of each user. The research challenge is to find the optimal challenge level for all user drivers in a shared simulation space that (1) involves both controllable entities (“opponents”) and non-controllable entities (users) and (2) is highly dynamic, with dependencies between entities being created and destroyed in real time. We try to solve this problem by modeling the scenario as a distributed constraint optimization problem (DCOP). The main contribution of our paper is the application of a DCOP algorithm to such a new type of application scenario. We evaluate our approach by running scenarios both in terms of speed and optimality of the solutions proposed by the DCOP algorithm.