Previous work has shown that combining Model Predictive Control (MPC) with Reinforcement Learning (RL) can effectively address the challenges of trajectory planning and control in dynamic environments with multiple robots. However, a key limitation of these approaches is the difficulty of designing a reward function that captures the complexities of real-world scenarios. To address this challenge, we investigate the application of curriculum learning to RL algorithms, with a focus on reward curricula that adapt the reward function over time to simulate increasingly realistic scenarios. Our approach starts with a simple reward function and gradually introduces more complex and realistic elements. By doing so, we aim to simplify the learning process and improve the sample efficiency of RL algorithms in complex environments.