Search

Coordinating multiple agents with diverse tasks and changing goals without interference is a challenge. Multiagent reinforcement learning (MARL) aims to develop effective communication and joint policies using group learning. Some of the previous approaches required each agent to maintain a set of networks independently, resulting in no consideration of interactions. Joint communication work causes agents receiving information unrelated to their own tasks. Currently, agents with different task divisions are often grouped by action tendency, but this can lead to poor dynamic grouping. This article presents a two-phase solution for multiple agents, addressing these issues. The first phase develops heterogeneous agent communication joint policies using a group communication MARL framework (GCMA). The framework employs a periodic grouping strategy, reducing exploration and communication redundancy by dynamically assigning agent group hidden features through hypernetwork and graph communication. The scheme efficiently utilizes resources for adapting to multiple similar tasks. In the second phase, each agent's policy network is distilled into a generalized simple network, adapting to similar tasks with varying quantities and sizes. GCMA is tested in complex environments, such as StarCraft II and unmanned aerial vehicle (UAV) take-off, showing its well-performing for large-scale, coordinated tasks. It shows GCMA's effectiveness for solid generalization in multitask tests with simulated pedestrians.

Even though the traffic systems in developed countries have been analyzed rigorously and operated efficiently, the same does not generally hold for developing countries due to inadequate planning, design, and operations of their transportation systems. Because of inherent differences between internal infrastructures, the strategies deployed in developed countries may not be amenable to developing ones. Besides, developing countries’ traffic systems are not well-studied in the literature to the best of our knowledge. For example, it is yet to explore how a developed country’s lane-based traffic flow would perform in the context of a developing country, which generally experiences non-lane-based traffic. As such, by using our newly developed traffic simulator ‘RoadBird,’ we investigate outcomes of both lane-based and non-lane-based traffic from the contexts of both developing and developed countries. To do so, we run simulations over real road topologies (extracted from the GIS maps of major cities such as Dhaka, Miami, and Riyadh), considering different scenarios such as lane-based or non-lane-based flows, homogeneous or heterogeneous traffic, with or without pedestrians, etc. We also incorporate various car-following and lane-changing models to mimic traffic behaviors and investigate their performances. While the lane changing dilemma remains an open research question, our experimental evidence indicates: 1) lane-based approaches will not necessarily perform better in the case of currently-adopted non-lane-based scenarios, and 2) non-lane-based strategies may benefit system performance in lane-based scenarios while having heavy mixed traffic. Nonetheless, we reveal several new insights for on-road experiences both in developing and developed countries.