Model-Agnostic Multi-Agent Perception Framework

Abstract

Existing multi-agent perception systems assume that every agent utilizes the same models with identical parameters and architecture, which is often impractical in the real world. The significant performance boost brought by the multi-agent system can be degraded dramatically when the perception models are noticeably different. In this work, we propose a model-agnostic multi-agent framework to reduce the negative effect caused by model discrepancies and maintain confidentiality. Specifically, we consider the perception heterogeneity between agents by integrating a novel uncertainty calibrator which can eliminate the bias among agents' predicted confidence scores. Each agent performs such calibration independently on a standard public database, and therefore the intellectual property can be protected. To further refine the detection accuracy, we also propose a new algorithm called Promotion-Suppression Aggregation (PSA) that considers not only the confidence score of proposals but also the spatial agreement of their neighbors. Our experiments emphasize the necessity of model calibration across different agents, and the results show that our proposed approach outperforms the stateof-the-art baseline methods for 3D object detection on the open OPV2V dataset.

Artificial Intelligence and Its Roles in the R&D of Vehicle Powertrain Products

“…Distinguished from the control of series or parallel HEVs, the control of the engine, generator, and traction motor cannot be coupled, and multi-inputs-multi-outputs (MIMO) control is required. This new requirement motivates the evolution from single-agent learning to multi-agent learning because the multi-agent system offers a feasible path for the MIMO control [50]. MADRL emphasizes the behaviors of multiple learning agents coexisting in a common environment with different collaboration modes.…”

Collaborative Perception in Autonomous Driving: Methods, Datasets and Challenges

“…Therefore, it is of great practical significance to ensure robustness and safety for collaboration perception. In this subsection, we focus on advanced research on these issues and summarize collaborative modules designed to address these issues [21,24,31,53,74,85,86].…”

“…When distinct agents are equipped with perception models in different architectures and parameters, current collaboration methods may generate unreliable fusion results due to model heterogeneity. To alleviate this issue, Chen et al [86] propose a model-agnostic collaborative perception framework. Firstly, considering there is a confidence distribution among different agents, an offline calibrator is used to align the confidence score of agents to its empirical accuracy.…”

“…Firstly, considering there is a confidence distribution among different agents, an offline calibrator is used to align the confidence score of agents to its empirical accuracy. Additionally, to conduct collaboration with the spatial correlation, [86] presents Promotion-Suppression Aggregation (PSA) module, which leverages an Intersection-over-Union (IOU) graph to find promotion proposals. The whole process is safe because the structure and parameters of the models are not touched.…”

Multi-vehicle Platoon Overtaking Using NoisyNet Multi-Agent Deep Q-Learning Network

“…By using advanced Vehicle-to-Vehicle communication technologies, AVs are able to share information with each other and cooperate in dynamic driving tasks. Through sharing information about the environment [1,2], locations, and actions, it will improve the driving safety [3], reduce the traffic congestion [4], and decrease the energy consumption [5]. Multi-vehicle collaboration and overtaking are two important topics for AVs.…”