Autonomous mobile robots have been used in outdoor delivery services. Delivery robots have to cope with dynamic obstacles and various environmental conditions. Although several successful technological solutions are available for indoor applications, there are still plenty of unsolved problems in outdoor environments. In this study, we concentrate on three technological challenges hindering the development of campus delivery robots. The first challenge is robust localization in various dynamic outdoor environments. Localization results obtained using lidars and Global Navigation Satellite System (GNSS) sensors show complementary advantages and disadvantages. The proposed localization strategy efficiently combines information from multiple sensors. The second challenge is safe navigation based on the detection of the traversable region. The presented terrain traversability analysis provides clear, real-time positive and negative obstacle information. The third challenge is the requirement of an effective path planning strategy to reduce the risk of collisions or deadlocks. We collected the appearance history of regional dynamic obstacles in the map. The resultant path was carefully generated to avoid crowded areas. Experimental verifications have been successfully conducted on the Korea University campus. The presented results clearly indicate that the proposed methods are essential to achieve safe and reliable navigation in dynamic, real-world environments.
Recently, HD maps have become important parts of autonomous driving, from localization to perception and path planning. For the practical application of HD maps, it is significant to regularly update environmental changes in HD maps. Conventional approaches require expensive mobile mapping systems and considerable manual work by experts, making it difficult to achieve frequent map updates. In this paper, we show how frequent and automatic updates of lane marking in HD maps are made possible with enormous crowdsourced data. Crowdsourced data is acquired from onboard low-cost sensing devices installed on many city buses and taxis in Seoul, South Korea. A large amount of crowdsourced data is daily accumulated on the server. The quality of sensor measurement is not very high due to the limited performance of low-cost devices. Therefore, the technical challenge is to overcome the uncertainty of the crowdsourced data. Appropriately filtering out a large amount of low-quality data is a significant problem. The proposed HD map update strategy comprises several processing steps including pose correction, observation assignment, observation clustering, and landmark classification. The proposed HD map update strategy is experimentally verified using crowdsourced data. If the changed environments are successfully extracted, then precisely updated HD maps are generated.
16 December 2024 in IEEE Robotics and Automation Letters
Optimizing large-scale logistics is computationally challenging due to its scale and requirement to be robust to stochastic and time-varying weather disturbances. However, prior research in multi-agent reinforcement learning (MARL) does not address scenarios that capture complexity of logistics operations influenced by dynamic weather patterns. To address this gap, we suggest a new MARL environment, MARINE that has two types of agents equipped with limited resources and integrates real wave data to model the influences of weather on the replenishment at sea (RAS) operation. To this end, we propose SchedHGNN, a novel MARL algorithm that incorporates a heterogeneous graph neural network and an intrinsic reward scheme to enhance agent coordination and mitigate challenges induced by environment non-stationarity. Our results show that the combination of effective RAS scheduling and improved communication enables our model to outperform competitive baselines by up to 37.8%. This achievement marks a significant advancement in applying MARL to complex, real-world logistics scenarios.