Page 25 - Read Online

Page 470 Phadke et al. Intell Robot 2023;3:453-78 https://dx.doi.org/10.20517/ir.2023.27

Table 6. Categorization of referenced study by the major resilience module/component they consider (general)
Resilience component/module highlighted Referenced study
Path planning [115]
Obstacle avoidance and detection [115]
Task assignment/reassignment [108,118]
Flocking [116,117]
Network connectivity and structure [116,119]
Swarm size scalability [116]
Agent property (heterogeneity) [108]
Scalable resilience evaluation metrics [120]
Navigation [110,112-114,121]
Area coverage [98]
Formation control [106,107]

property component. An application where budget constraints apply is for efficient sensing of the
[109]
environment using low-cost sensors onboard toy drones. The study in uses a divided framework to map
its surrounding terrain with a fast 3D model in the frontend and an offline backend that uses an MVS
(Multi-View Stereo) to create a higher resolution 3D model. Such models aim to combine fast sensing and
data acquisition platforms with post-processing methodology for high-resolution data acquisition. This
provides a good example of using fast MVS methodologies to create an automated low-cost 3D map from
inexpensive UAVs, with potential applications in surveying and mapping. For accurate comprehensive
mapping ability, an optimized area coverage approach is vital. Most Flying Ad Hoc Network (FANET)
applications require the ROI (Region of Interest) to be maximally covered, with additional constraints such
as time and resource limitations. Setting optimized waypoints is often the first step in semi-autonomous
[98]
flight planning systems . This, combined with dynamic grid decomposition and selection schemes,
provides better results than planner-only approaches.

Autonomous swarm navigation with multi-target sensing and tracking in the presence of dynamic obstacles
is another such generalized development that can be applied to military, SAR, and general surveying
applications discussed above. Navigation methods need to be fault-tolerant and capable of functioning in
sparse and GNSS-deprived environments [110,111] . Indoor localization [112,113] is an additional challenge where
GNSS signals might be weaker, hence relying on other sensor readings, passive beacons, fiducial
[114]
markers , or cooperative localization techniques.

Article discusses a multi-view approach to swarm management, path planning, and obstacle detection
[115]
and tracking using trust region policy and proximal policy optimization algorithms. An interesting
assumption by the study is that the agents are homogeneous; heterogeneous agent additions have been
[116]
omitted. The authors in present a distributed flocking protocol for mid-sized UAV swarms (< 100
agents) where they define swarm size, communication radius, and collision parameters to create resilient
implementations and a research methodology for future development in control theories and statistical
analysis of the results. This requires the design of relevant scalable metrics for performance evaluation.

Data transfer policies between swarm agents and non-connected swarm entities are a network issue for
[117]
swarm devices. Article proposes an adaptive data transfer method for separated non-swarm devices using
offline evolving swarms to enable the connection between disconnected network nodes. The resultant
swarm was able to adapt emergent behavior and achieve effective transmission between the desired nodes.
Resource allocation and task assignment problems are also prevalent issues in all scenarios, application-