Fig. 3. Example of condensed compaction report with undesired densities after compaction (Area I: Inhomogeneous compaction, Area II: Partially insufficient compaction, Area III: Insufficient compaction, Area IV: Insufficient compaction).

24 R. Kuenzel et al. / Automation in Construction 71 (2016) 21–33

Table 1

Related work in the field of assistance systems for compactor operators.

Authors Year Evaluation method Quantitative results Difference to our approach

Utsuka et al. 1992 Field test Not reported Predefined working area instead of a dynamically changing area;

environmental conditions (e.g. temperature) not considered.

Froumentin & Peyret 1996 Field test Not reported Environmental conditions (e.g. temperature) are not considered.

Fujikawa et al. [15]

1996 Field test Avg. & stdv. of reached density Environmental conditions (e.g. temperature) are not considered.

Tserng et al. 1996 Simulation Not reported Environmental dynamics and temporal restrictions were not considered.

Tserng et al. 1996 Informed argument Not reported Fixed rolling patterns and paths. No real-time sensory inputs; no

real-time decisions about paths.

Lee et al. [16]

1998 Simulation Math. function for density acc. to temp. & No automation of path planning; method of data gathering unclear

no. passes

Lee et al. 1998 Informed argument Not reported Fixed rolling patterns and paths. No real-time sensory inputs; no

real-time decisions about paths.

Peyret 2000 Field test Not reported Environmental conditions (e.g. temperature) are not considered.

Furuya & Fujiyama [17]

2011 Field test Distribution function for stiffness Sensory input is limited to stiffness.

Vasenev et al. [18]

2014 Field test Not reported No automation of path  planning.

2.3. Technical background

This paper adopts techniques from the field of multi-agent systems to automatically generate driving instructions for human compactor operators based on commercially available solutions for acquiring, com- municating, and storing real-time sensory inputs. A software agent “is a computer system that is situated in some environment, and that is capa- ble of autonomous action in this environment in order to meet its design objectives” [19]. Intelligent agents hold a model of their environment (as their “mental state”) continuously updated by sensory perceptions. Intelligent agents are realized as intentional artefacts [20] described by a knowledge base (e.g. rules for generating rolling patterns, physical cause–effect-models) and intentions to reach a certain goal state like a desired asphalt density after compaction. Agents are able to communi- cate with each other in order to coordinate and cooperate; an agent is also able to perform action via its actuators. An action can be a manipu- lation of the physical world (e.g. open and close a valve) or it can be a virtual/digital action like sending a calculated driving instruction to the display of the compactor operator. Fig. 4 shows a graphical overview of an intentional software agent.