Cooperative Multi-Robot Control for Target Tracking with Efficient Switching of Onboard Sensing Topologies

Using multiple robots to track a moving target is potentially beneficial because of the reduction in tracking uncertainty, increased coverage, and robustness to failure. Two problems arise immediately. First, these objectives are often at odds (e.g., the configuration of the robots that lead to the lowest uncertainty estimates of target pose may not be the best if one or more robots is disabled). Second, the robots themselves are often poorly localized (e.g., only a few may have access to GPS, and the rest may be limited to a combination of onboard inertial sensing, visual odometry, and relative range/bearing measurements to estimate their poses relative to each other). In the domain of cooperative control, small unmanned aerial vehicles (UAVs) have recently become prominent in multi-robot control with motion capture state estimates [5, 7]. For cooperative target tracking with onboard sensors, researchers considered centralized [3], decentralized [1], and distributed [4] approaches to multi-robot control in aerial and ground settings. However, these methods estimate the pose of the target and assume that the poses of the robots are known, e.g., from an external system or by reference to a global map. To robustly perform cooperative multi-robot localization using only onboard sensors, optimization-based maximum-likelihood localization approaches have been proposed [2]. However, it does not allow for direct minimization of the uncertainty associated with the estimated target pose. In this paper, we consider the cooperative control of a team of robots to estimate the position and minimize the position uncertainty of a target using onboard sensing. In particular, we assume limited sensing capabilities and reason over the entire sensing topology by explicitly estimating the joint state of the robots and target.