An Extended Bayesian Optimization Approach to Decentralized Swarm Robotic Search

Abstract

Swarm robotic search aims at searching targets using a large number of collaborating simple mobile robots, with applications to search and rescue and hazard localization. In this regard, decentralized swarm systems are touted for their coverage scalability, time efficiency, and fault tolerance. To guide the behavior of such swarm systems, two broad classes of approaches are available, namely, nature-inspired swarm heuristics and multi-robotic search methods. However, the ability to simultaneously achieve efficient scalability and provide fundamental insights into the exhibited behavior (as opposed to exhibiting a black-box behavior) remains an open problem. To address this problem, this paper extends the underlying search approach in batch-Bayesian optimization to perform search with embodied swarm agents operating in a (simulated) physical 2D arena. Key contributions lie in (1) designing an acquisition function that not only balances exploration and exploitation across the swarm but also allows modeling knowledge extraction over trajectories and (2) developing its distributed implementation to allow asynchronous task inference and path planning by the swarm robots. The resulting collective informative path planning approach is tested on target-search case studies of varying complexity, where the target produces a spatially varying (measurable) signal. Notably, superior performance, in terms of mission completion efficiency, is observed compared to exhaustive search and random walk baselines as well as a swarm optimization-based state-of-the-art method. Favorable scalability characteristics are also demonstrated.

Active Sensing for Search and Tracking: A Review

“…1) Challenges in task-driven control: To increase the flexibility of pure-exploitative strategies, hybrid solutions can be adopted, even though in hybrid control the performance strongly depends on accurate exploitation-exploration balancing methods [16]: if the controller is biased towards the source-seeking term, the estimation process might get stuck on local optima and the target may never be found; on the other hand, higher importance on the exploration term, leads to superior capabilities to explore the environment and, therefore, to find the target. However this may come at the cost of higher completion times, sometimes not compatible with the mission requirements [16]. In conclusion, hyperparameter tuning is not only cumbersome and time-consuming, but also critical for the performance of hybrid APE solutions.…”

“…Probabilistic approaches account for realistic perception uncertainties [83]; hence, they are suitable to manage real-world (noisy) scenarios, unmodeled dynamics, sensing nuisance. Furthermore, probabilistic decision making has high adaptivity properties [13], and it is useful when poor a-priori knowledge is available [59] Finally, the stochastic formulation of the measurement process allows to model and account for the lack of data through the negative likelihoods (16).…”

“…Some APE problems are formulated as environmental mapping tasks, where the target location is inferred from the estimated spatial distribution of a physical quantity (e.g., source signal strength [16], [74]) or from occupancy maps [70], [127]. In this framework, a common approach is to tesselate the workspace Π into a grid map and apply RBE to each cell of the grid [127].…”

“…The acquisition function can also be a linear combination between a source seeking term and an 13 MMSE: Minimum Mean Square Error. 14 When the platform position is fixed and only rotation movements are allowed (e.g., pan cameras on a static robot), the distance is computed between the setpoint and the centre of the FoV projection onto the groundplane Π [215] information gathering one [16]; the former might account for the distance to the maximum expected target position, while the latter should depend on the target belief uncertainty.…”

Learning Robot Swarm Tactics over Complex Adversarial Environments

“…We refer to the single behaviors of a group as a primitive. Examples could be formation control [24], distributed mapping [25], signal source localization [26], group coverage [27], and others. Tactics are ensembles of primitives commanded on multiple groups for swarm behavior.…”