This article presents the development of a mobile robotic system based on an 18-degree-of-freedom hexapod, designed to autonomously collect and classify waste-like objects in a controlled environment. The project incorporates a computer vision system using a high-resolution overhead camera, whose processed images enables real-time identification of both, the robot’s position and the location of objects distributed in the workspace. Trajectory planning was implemented through the Probabilistic Roadmap (PRM) navigation algorithm, developed in Python and integrated into ROS 2 Humble. Locomotion control was structured as a decision-making algorithm that, combined with predefined walking routines, ensures navigation toward target points. For object manipulation, a flexible gripper was designed following soft robotics principles, manufactured additively manufacturing and optimized to adapt to diverse object geometries. The system was validated in both simulated environments using CoppeliaSim and real experiments, demonstrating its ability to detect objects, generate trajectories, perform manipulations, and transport items to designated drop-off zones. This work contributes a modular ROS 2 architecture for legged robots that combines navigation and soft manipulation, validated in both simulation and real-world experiments.