NetBots is a novel swarm robotics algorithm to improve detection accuracy in the distributed automatic target recognition (ATR) problems. Prior work (Dasgupta et al. and Martinoli et al.) have applied swarm robotics to the ATR problem by having a large number of cheap, inaccurate robots "vote" on whether a given object is a target.
These prior works, however, assume that the host of inexpensive robots is homogeneous in detection accuracy (i.e. all robots have the same probability of misidentifying a target). This is an inaccurate assumption: inexpensive production techniques are error-prone and produce robots of widely varying accuracy. The NetBots algorithm uses a decentralized social network of robot interactions to approximate each robots' individual accuracy. With this additional information, the robots' votes can be weighted to give more accurate robots greater influence when classifying a potential target.
Both the NetBots and the HelpBots (the naïve control algorithm) were simulated in the open-source Stage simulator using custom C++ robot controllers. Initial results show a significant decrease in the number of errors experienced by the NetBots algorithm when compared to the control HelpBots algorithm. This research was submitted to the IEEE ICRA 2010 conference, but was not selected to be published.
This work was supported in part by the NSF Grants: 0341601, 0647018, 0717674, 0717680, 0647120, 0525429, 0806931, 0837332.