Monitoring coastal fish assemblages through visual census is widely adopted, yet affected by observer-dependent variability and logistical constraints. This study provides a preliminary assessment of the feasibility of using an autonomous surface vehicle (ASV) with omnidirectional navigation for non-invasive monitoring of coastal teleost assemblages. The system was tested on Mugilidae and Sparidae along the southern Tyrrhenian coast of Calabria. Surveys were conducted in two ecologically distinct systems: a coastal marine area (Paola (CS), origin of inertial frame 39° 19' 11.80" N, 16° 2' 34.38" E) and a brackish lagoon environment (Laghi di Gizzeria (CZ), origin of inertial frame 38°56'4.19'' N, 16°12'29.09'' E). For each site, three 50 m linear transects were performed per season, recording presence/absence and abundance on rocky and sandy substrates. Overall, 192 individuals belonging to six taxa were observed. The marine site exhibited higher species richness, with five species and 138 individuals, whereas the brackish system was exclusively dominated by Chelon labrosus (Risso, 1827), with 54 individuals distributed across seasons. The autonomous platform enabled transect execution at constant speed and predefined spatial parameters. Electric thrusters were configured to ensure over-actuated propulsion and low acoustic impact, minimizing disturbance and maintaining non-invasive monitoring conditions. The observed patterns were consistent with the ecological characteristics of both environments, supporting operational feasibility under the tested field conditions. Although based on preliminary data, the approach demonstrates methodological consistency across repeated surveys under different coastal environmental conditions and potential applicability to different components of coastal teleost assemblages. Designed to work alongside human operators, the system may complement monitoring by supporting standardization and extending observational capacity in a continuous, non-invasive, and replicable manner. A thorough comparison with conventional visual census or other validated monitoring approaches will be required to quantify any potential reduction in observer-dependent variability and assess comparative monitoring reliability.