a b s t r a c t  ：This article presents an overview of the SmartSite research project that adopts machine learning, decision theory and distributed artificial intelligence to design and test a multi-agent system (MAS) for asphalt road construction. SmartSite puts major emphasis on sensing and communication technologies that integrate real-time automated information exchange in the supply chain of road construction. As part of the larger SmartSite project, this article introduces a novel real-time path planning system for compactors and presents the results of several simulation and field realistic experiments conducted to evaluate the system in a sophisticated simulation and harsh con- struction environment, respectively. The system operates based on Belief-Desire-Intention (BDI) software agents and real-time sensory inputs. The newly developed integrated and information rich process benefits asphalt compactor operators, as they are now capable to control their machinery and react to changing environmental, material-related and process-related disturbances or changes. This improves the quality of the delivery and laying of asphalt material, prevents compactors from over-compacting certain road segments, increases the road's pavement longevity during the operational life cycle phase; refocuses the work tasks of the site managers, and reduces the construction budget and schedule. The system's ability to maneuver an asphalt roller during real- word operation also makes it an important step towards a fully automated asphalt compactor.

1. Introduction

The European construction industry contributes nearly 10% to the EU-27-GDP [1] but close to 90% of its projects suffer from budget overruns or delays [2]. A main cause seems to be inadequate planning and management, in particular of capital-intensive infrastructure construction projects.

Modern road construction operations are in unique and dynamic work environments. They require highly complex logistics systems and tight coordination among the multiple participating stakeholders, including companies with vastly different economic objectives, inter- as well as intra-procedural dependencies. Current manual project coordination methods fail in addressing these issues adequately. As a result, project budget overruns and delays are quite common.

Recently, several techniques emerged to support project site management with in-time information about the construction process- es and performance of machines and staff. Multi-agent systems (MAS) tend to be a promising technology for automated site  management

that reduces the centralistic, strictly hierarchical work portfolio of a tra- ditional site manager.

In the last phase of black-top assembly, asphalt compaction is one of the most important steps in road construction. Faults during this phase are almost always irreversible without incurring high costs for milling- off and re-building [3]. Construction experts estimate that 5% of the con- struction sum is spent on fixing quality defects. In Germany alone, 2.2 B

€ are spent annually to mitigate quality defects on public roads [3]. An adequate compaction of asphalt is crucial in high quality road construc- tion and for the longevity of the roads.

Key determining factors for the compaction process are: environ- mental preconditions, like ambient air temperature, wind speed or the amount of rainfall, and material properties of the surface like thermal conductivity or core temperature of the asphalt [4]. These influencing factors are not under the control of the compactor oper- ators (see Fig. 1).

Asphalt density – and therefore quality – is essentially dependent on machine handling and driving behavior of the compactor operators. Operators have to judge about the machine velocity, number and uniformity of passes of certain areas, overlapping areas and vibration parameters. Thus, operators need to be very sensitive towards the uniformity of compacting alongside the whole pavement mat. The latter is determined according to numerous environmental, material-related and process-related pre-conditions.