Plastic pollution in aquatic environments is a serious ecological concern that poses a significant threat to marine life and has a detrimental effect on ecosystems. Autonomous surface vehicles (ASVs) offer a scalable and safe solution for environmental cleanup missions, particularly in dynamic and partially observable aquatic environments. By decoupling the cleanup missions into an initial exploration phase followed by a cleaning phase, this chapter addresses the challenge of locating highly dynamic trash items. This chapter introduces a multitask multiagent deep reinforcement learning (MT-MADRL) framework to optimize ASV coordination for both exploration and waste collection, enhancing scalability by enabling knowledge transfer across multiple agents using a single policy trained using deep reinforcement learning (DRL). The proposed method is evaluated across multiple cleanup scenarios, demonstrating its ability to efficiently adapt to dynamic environments. Additionally, a Pareto front analysis is constructed to evaluate trade-offs between two competing objectives, such as exploration and cleaning.