The field of multi-agent systems and autonomous driving is rapidly evolving, with a focus on improving perception, prediction, and decision-making in complex scenarios. Recent developments have seen a shift towards more efficient and effective methods for collaborative perception, multi-agent simulation, and agent modeling. The use of transformer-based architectures and probabilistic frameworks has shown promising results in enhancing the accuracy and robustness of these systems. Furthermore, the integration of simulated data and real-world scenarios has improved the performance of autonomous driving systems in challenging situations. Noteworthy papers in this area include CoST, which proposes an efficient collaborative perception method that aggregates observations from different agents and times into a unified spatio-temporal space, and RoboTron-Sim, which improves real-world driving in critical situations by utilizing simulated hard cases. Additionally, papers such as TransAM and MIDAR have made significant contributions to agent modeling and LiDAR detection mimicking, respectively.
Advancements in Multi-Agent Systems and Autonomous Driving
Sources
CoST: Efficient Collaborative Perception From Unified Spatiotemporal Perspective
On Learning Closed-Loop Probabilistic Multi-Agent Simulator
TransAM: Transformer-Based Agent Modeling for Multi-Agent Systems via Local Trajectory Encoding
MIDAR: Mimicking LiDAR Detection for Traffic Applications with a Lightweight Plug-and-Play Model
RoboTron-Sim: Improving Real-World Driving via Simulated Hard-Case
TurboTrain: Towards Efficient and Balanced Multi-Task Learning for Multi-Agent Perception and Prediction
Driver Assistant: Persuading Drivers to Adjust Secondary Tasks Using Large Language Models
DistillDrive: End-to-End Multi-Mode Autonomous Driving Distillation by Isomorphic Hetero-Source Planning Model