Autonomous robots are transforming the agricultural industry by integrating automation into the inspection of crops and plants. Primarily, this enhances production through periodic spatio-temporal assessments, assists in optimizing resource allocation and evaluates the effects of climate change via uninterrupted surveys. These surveys require a smart navigation strategy, such as the lawnmower pattern, where ground robots can collect high-precision images and sensory information without interruption. In this work, reinforcement learning-based frameworks are explored to enable collision-free and constrained navigation for collecting spatio-temporal data in a vineyard environment. This approach formulates a pose correction strategy integrated with visual feedback by deploying Soft Actor-Critic (SAC) and Proximal Policy Optimization (PPO) algorithms. A custom reward mechanism is introduced to constrain the robot’s pose within a ±25° threshold for every 1.5-meter longitudinal distance by optimizing the waypoint coefficient and orientation coefficient, respectively. Furthermore, this optimization restricts lateral movement to a range of 0.025–0.125 meters for SAC and 0.020–0.175 meters for PPO.