Title: Multi-Agent Reinforcement Learning for Autonomous Mobility on Demand.
Abstract: This talk discusses the use of multi-agent reinforcement learning (MARL) for autonomous mobility on demand applications. The research is focused on designing MARLs for homogeneous agents that are capable of taking orders, recharging, and bidding for orders. The objective is to reduce customer wait time, empty miles, and increase achieved orders and fleet utilization. Furthermore, the study investigates the impact of experience sharing between agents on improving the learning phase. Preliminary results indicate a significant improvement when two agents are connected in the exploitation phase compared to fully independent learners. Additionally, a further improvement is observed when more agents share their experience in a broadcast manner. Future investigations will focus on multi-objective reward design for the same context, with the aim of optimizing multiple metrics simultaneously, such as customer satisfaction, fleet utilization, and profitability.
Title: Robust 3D Object Detection for Autonomous Vehicles with Cross-Modal Hallucination
Abstract: This talk will present a novel point-based framework for robust 3D object detection that utilizes the complementariness of LiDAR and 4D radar sensors, with inference using only one modality. The proposed method addresses the limitations of each sensor. It introduces an instance feature aggregation module and a feature alignment step to deal with geometry discrepancies for better cross-modal instance matching and intrinsic attribute gaps between the sensing modalities. Experimental results on the View-of-Delft dataset demonstrate that the proposed method outperforms state-of-the-art radar and LiDAR object detection methods while maintaining competitive efficiency in runtime.