Back to search

Ελληνικό Μεσογειακό Πανεπιστήμιο

Intelligent actuation and resource-efficient localization and navigation of vision-based autonomous robotic systems

Abstract

dc:description

Robots are most needed in the environments where satellite positioning is unavailable: building interiors, underground workings, pipe networks and disaster sites. In such settings a vehicle must estimate its own pose from onboard sensing alone and plan safe motion without any prior knowledge of the environment. The two problems are coupled. The quality of the pose estimate bounds what can be planned and executed safely, while the structure of the environment determines what observations the estimator has to work with. This dissertation addresses localization and local navigation jointly, across three settings that stress the coupling differently: cluttered indoor spaces containing moving obstacles, outdoor disaster scenarios requiring several coordinated vehicles, and confined pipe interiors that offer almost no visual structure. A survey of SLAM and odometry organized by sensing configuration first establishes the state of the art across six sensor combinations and four maturity tiers, and identifies the open problems the remaining work addresses. A local navigation framework is then developed that couples an incrementally built Euclidean signed distance field with an optimization-based B-spline trajectory replanning, maintaining the map so that it reflects the current rather than the historical position of obstacles, and is validated on aerial and ground vehicles in simulation and on a skid-steering platform. This framework is extended with a velocity-adaptive map whose size and resolution scale with the vehicle's motion state at approximately constant voxel count, combined with an angular-velocity-based lateral offset, and a short-term vicinity memory. Implemented on a differential-drive vehicle developed for this purpose and benchmarked against three state-of-the-art navigation systems, it succeeds in geometrically demanding configurations where search-initialized methods fail in every trial. A further extension coordinates a multi-agent aerial swarm by adding a formation-similarity term based on the normalized graph Laplacian and a reciprocal avoidance term to the cost function of the optimization framework, requiring neither a leader-follower hierarchy nor a global map. Localization is finally addressed in pipe interiors, where the estimator is deprived of usable observations rather than merely inaccurate. Trackable structure is generated actively by depositing non-permanent markings onto the pipe wall. A closed-form analysis of the Fisher information for specific force shows that the information is maximized when the inertial sensor positions satisfy a centroid condition, realized by a uniformly spaced ring. The resulting eigenvalue structure provides greater information along the rotational axis that dominates in-pipe motion, while the optimality result is invariant to known per-sensor orientations. Two conclusions follow. First, a local distance-field representation coupled with optimization-based replanning transfers across platforms, dynamics, and mission structures, with adaptation confined to the cost function. Second, when the environment defeats the available sensing, acting on the environment can be more consequential than further refinement of the estimator.

Degree

thesis:*
Grantor dc:publisher
Ελληνικό Μεσογειακό Πανεπιστήμιο
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Geladaris, Aristeidis-Stavros
  • Γελαδάρης, Αριστείδης-Σταύρος

Subjects

dc:subject × 18

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62746

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Geladaris, Aristeidis-Stavros; Γελαδάρης, Αριστείδης-Σταύρος. Intelligent actuation and resource-efficient localization and navigation of vision-based autonomous robotic systems. Ελληνικό Μεσογειακό Πανεπιστήμιο, 2026. http://hdl.handle.net/10442/hedi/62746