Tham khảo tài liệu 'multi-robot systems from swarms to intelligent automata - parker et al (eds) part 3', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | Sensor Network-Mediated Multi-Robot Task Allocation 35 bars and hence is favored for use in this application since it provides reduced bounds on system run time over a simple Raster Scan method. We also compared mobility requirements for DINTA and RS methods. Specifically the use of mobility requires energy. A measure of energy for mobility is determined for the purposes of comparison by computing the total time of the robot motion. Figure 4 shows a comparison of energy consumption in units of time-in-motion. As expected DINTA outperforms Raster Scan significantly. However as the number of events increases to infinity DINTA will approach Raster Scan energy consumption. Also note that on the interval 5 20 the slope of the Raster Scan curve is very small and the energy consumption is insensitive to event arrival rate. 6. Field Trials using NIMS The third and final set of experiments discussed here were performed in field trials with the NIMS system. We used our task allocation system and compared two policies - Time tasks with smaller time stamp get priority and Distance tasks closer to the robot get priority . A set of experiments was conducted on a NIMS setup deployed in the James San Jacinto Mountain Reserve. Because of space limitations only representative graphs are presented. Figure 5 shows the representative PAR data from sensor 1 collected during the operation of the Time policy Figure 5a and the Distance policy Figure 5b . Figure 5 also shows points in time when events were generated and serviced by both policies for sensor 1. Note that events are generated in response to fluctuations in PAR. As shown on Figure 5 events are generated proportionally to the density of the spikes in PAR data and cover all significant spikes of PAR data. Figure 5c shows the comparison between the cumulative event OnTime of the Time policy and the Distance policy. For visualization purposes in Figure 5c event s OnTime is presented as a zero-mean Gaussian distribution. It follows
