Active Acoustic Source Tracking Exploiting Particle Filtering and Monte Carlo Tree Search

Abstract

In this paper, we address the task of active acoustic source tracking as part of robotic path planning. It denotes the planning of sequences of robotic movements to enhance tracking results of acoustic sources, e.g., talking humans, by fusing observations from multiple positions. Essentially, two strategies are possible: short-term planning, which results in greedy behavior, and long-term planning, which considers a sequence of possible future movements of the robot and the source. Here, we focus on the second method as it might improve tracking performance compared to greedy behavior and propose a path planning algorithm which exploits Monte Carlo Tree Search (MCTS) and particle filtering, based on a reward motivated by informationtheoretic considerations. By representing the state posterior by weighted particles, we are capable of modelling arbitrary probability density functions (PDF)s and dealing with highly nonlinear state-space models.

Active Sensing for Search and Tracking: A Review

“…[65] AL static P 1 > 1 [8], [43] AT dynamic P 1 1 [18], [66], [67] AT dynamic A 1 1 [40], [68] AT coordinates [20]). To this aim, target and (robotic) sensing platforms may establish some form of direct communication, for instance via radio [18] or acoustic [53] devices.…”

“…hazardous chemicals, fluids, or gaseous particles [22]). The presence of active targets is usually captured by robots equipped with ad-hoc sensing capabilities (i.e., radio receivers [69], microphone arrays [66], gas detectors [24]); once the target is perceived, its position is inferred through the extraction of specific features from the collected data (e.g., the signal strength in radio communication [73]) Active targets often carry electronic devices that send information to the APE system (e.g., portable WiFi transmitters [74]). Despite this, it is important to remark that a target is not required to carry an emitting device to be active: it can be a device-free speaker [35] or a radiation source in critical environments [75].…”

“…Active Localization (AL) and Active Tracking (AT) use measurements provided by single [66] or multiple [40] onboard sensors to control the platform while estimating the target position. Sensing modalities include vision [2], radar [77], audio [66] and radio signals [37].…”

“…Active Localization (AL) and Active Tracking (AT) use measurements provided by single [66] or multiple [40] onboard sensors to control the platform while estimating the target position. Sensing modalities include vision [2], radar [77], audio [66] and radio signals [37]. Measurements may be aggregated through sensor fusion, which can be uni-modal [40] or multi-modal [35], cooperative [24] or non-cooperative [37].…”

“…MUSIC [133] or SRP-PHAT [134] algorithms are typically used to estimate the relative Angle of Arrival (AoA) between the robot and the source. Alternatively, AoAs can also be measured sequentially, for instance by a moving robot equipped with microphone arrays [66]. On the other hand, inferring distance information from microphone data is generally difficult in reverberant rooms and often relies on labeled training data [135].…”