The importance of transportation cannot be overstated, with road maintenance and construction being among the most crucial sectors. However, this area has been slow to update its tools and procedures, despite the benefits of automation. By embracing automation, the road construction industry can realize benefits such as increased efficiency, reduced physical strain on workers, shorter construction times, and less economic loss. In the road construction environment, traffic cones are commonly used to delimit work areas. These cones must be placed by workers and moved as the project progresses. Automation can greatly accelerate this process, freeing up workers for more complex tasks. However, conventional robots require an operator to control the device, limiting the efficiency gains.

To address this inefficiency, we propose a solution based on a robot that can autonomously reach the desired position. Our objective is to develop a model of a robotic cone using reinforcement learning, enabling it to operate independently and improve the efficiency of road construction projects. The self-learning is based on a system of rewards and punishments to achieve the desired position. The cone is rewarded if it approaches or reaches the goal, but it is penalized if it moves away, exceeds the goal or is exploring a wrong quadrant. By using this method, the cone must choose between a 0º or 90º each step-time to maximize the long-term reward. The simulated robotic safety cones reach the target, but the large number of variables involved long training times.