Feeding assistance is a fundamental challenge in assistive robotics, requiring robust perception, adaptive motion control, and safe human–robot interaction. This paper presents a preliminary framework for autonomous robotic feeding, structured into three modules: utensil grasping, bite acquisition, and food transfer. The framework integrates visual perception with Learning from Demonstration (LfD), using Gaussian Mix- ture Models (GMM) for trajectory adaptation. Utensil grasping is achieved through affordance detection, food segmentation is performed using HSV-based color detection, and mouth local-ization is estimated with Mediapipe. Experiments conducted with a 6-DoF manipulator and RGB-D cameras show an overall success rate of 65% across complete feeding trials, with the highest reliability in utensil grasping (95%) and food transfer (85% given successful bite acquisition). Results highlight the bite acquisition stage as the main bottleneck due to limitations of color-based segmentation. Future work will address these challenges by integrating foundation models for robust food detection, expanding food variety, and incorporating user-in the-loop interaction for natural bite selection.