Optimizing Consensus-based Multi-target Tracking with Multiagent Rollout Control Policies

Abstract

Tracking multiple moving objects of interest (OOI) with multiple robot systems (MRS) has been addressed by active sensing that maintains a shared belief of OOIs and plans the motion of robots to maximize the information quality. Mobility of robots enables the behavior of pursuing better visibility, which is constrained by sensor field of view (FoV) and occlusion objects. We first extend prior work to detect, maintain and share occlusion information explicitly, allowing us to generate occlusion-aware planning even if à priori semantic occlusion information is unavailable. The efficacy of active sensing approaches is often evaluated according to estimation error and information gain metrics. However, these metrics do not directly explain the level of cooperative behavior engendered by the active sensing algorithms. Next, we extract different emergent cooperative behaviors that stem from the same underlying algorithms but manifest differently under differing scenarios. In particular, we highlight and demonstrate three emergent behavior patterns in active sensing MRS: (i) Change of tracking responsibility between agents when tracking trajectories with divergent directions or due to a re-allocation of the resource among heterogeneous agents; (ii) Awareness of occlusions to a trajectory and temporal leave-and-return of the sensing agent; (iii) Sharing of local occlusion objects in MRS that subsequently improves the awareness of occlusion.

SMA-NBO: A Sequential Multi-Agent Planning with Nominal Belief-State Optimization in Target Tracking

“…The robot team leverages communications capability to operate in a distributed manner, with each agent i processing local observation Z i,k at time instant k and generating local information, then making average-consensus on target estimation with its neighbors in a fixed number L of consensus steps. The data association and consensus component is consistent with the description in [9]. After the agents' local processing and consensus steps, the fleet provides the estimation of target in set Θ k at a constant sampling rate, ∆t (frequency f 1 = 1/∆t Hz).…”

“…A. An Information-Driven POMDP Similar to the work in [9], a POMDP model is formulated in this multi-agent scenario. Define a POMDP model as a tuple P = (X , O, U, T ,C).…”

“…In information theory, entropy is one solid mathematical quantification of information gathering and is treated as the objective to minimize via observation selection and action optimization [2], [5]. Another approach to target tracking seeks to plan the robot motion towards reduction of the mean squared error (MSE) of associated targets, which leads to the trace of covariance matrix tr [P] as the indicator of planning [6], [7], [8], [9].…”

“…To make such an optimization computationally tractable, future trajectories need to be approximated. The Monte Carlo method is studied as one solution to such approximation, which draws target trajectory samples from beliefs of the current target state and assumptions on the dynamic behavior of the target [9], [10]. The disadvantage of such an approach is the computational issue for general Monte Carlo methods since a significant number of samples are required for a good approximation.…”

“…The primary contributions of this paper are: (i) A computationally efficient multi-agent tracking algorithm that utilizes the policy of intent in sequential planning, extending works from [9] that uses belief state optimization on multi-agent tracking problem and [15] that uses NBO on a single agent tracking problem. (ii) The introduction of the concept of a random forest to automatically simulate a large number of occluded environments with different quality of occlusions, and use of the random forest to statistically demonstrate the superior performance of the proposed algorithm.…”